Technologies for evaluating health conditions and / or life stages

Abstract

Technologies are disclosed for evaluating an intervention effect in a condition for a plurality of subjects. A first data may be received from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. For each of the plurality of subjects, a second data may be received identifying one or more indicators of a life stage or a life-quality condition for which each of the plurality of subjects received the therapy, the nutrition, or the intervention over at least some portion of the period of time. The first and second data may be input into at least one machine-learning (ML) model that may process the first and second data. The ML model may determine the intervention is relatively effective, or that the intervention is relatively ineffective.

Description

TECHNOLOGIES FOR EVALUATING HEALTH CONDITIONS AND / OR LIFESTAGESCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 615,078, filed on December 27, 2023, the entire contents of which is hereby incorporated by reference herein, for all purposes.BACKGROUND

[0002] Dermatological conditions, for example, can impact both humans and animals, including pets. Dermatological conditions include acne, atopic dermatitis, shingles, hives, sunburn, contact dermatitis, rosacea, autoimmune diseases, basal cell carcinomas, bacterial infections, allergies, fungal infections, among others.

[0003] Dermatitis is a common dermatological condition. Dermatitis symptoms include itchy skin, dry skin, skin rash, hair loss, greasy and / or matted coat, flaking skin, thickened skin, and / or skin blisters. One or more intervention effects may remediate dermatitis symptoms, such as moisturizers, corticosteroids, antihistamines, specialized bandages, special diets, and / or prescription medications, among other intervention effects, and other therapies such as shampoos, lotions, and specialized foods such as therapeutic foods, elimination diets, supplements, or other specialized nutrition therapies may be needed to aid in the management of such conditions.BRIEF SUMMARY

[0004] Technologies are disclosed for evaluating a intervention effect and / or a change of a condition / life stage for a plurality of subjects. One or more processing devices may be configured to perform one or more techniques. One or more techniques may comprise receiving a first data from a plurality of activity monitors that may be respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time.

[0005] One or more techniques may comprise, for each of the plurality of subjects, receiving a second data that may identify one or more indicators of a life stage, and / or a life-quality condition, for which each of the plurality of subjects received the therapy, the nutrition, and / or the treatment over at least some portion of the period of time.

[0006] One or more techniques may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more techniques may include processing, by at least one ML model, the first data and the second data.

[0007] One or more techniques may comprise determining, via the at least one ML model, that the intervention effect is relatively effective, or that the intervention effect is relatively ineffective, for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data.

[0008] One or more techniques may comprise identifying (e.g., new) data features that are indicative of dermatological issues. These may be (e.g., novel) features that are not apparent / obvious to pet parents and / or veterinarians. These features may be combined in various ways to form a dermatological index (e.g., a composite that may represent one or more, or multiple, aspects of the dermatological condition, perhaps for example not just scratching), and / or to analyze the data using a responder analysis and / or to analyze the data using a responder analysis, a machine learning model, or some other type of analysis, for example.

[0009] In one or more scenarios, at least Some of the (e.g., new) inventive aspects include: 1) collecting data at a high level of resolution (ex., 50 Hz, 100 Hz, among other higher and lower resolutions), 2) analyzing the data using (e.g., proprietary) algorithms and / or any algorithms that are capable of labeling behavior (for example one or more algorithms may be used to label scratching, shaking, walking, running, resting, sleeping, and / or sleep quality), 3) using the labeled behaviors to generate many different features in the data,, , 4) considering features at the level of a day or at an even finer level of granularity, such as by quadrant (e.g., 6 hour increments) of the day, by hour, by minute, by second, or by sub-second, for example, to avoid the loss of potentially important information that may occur if data was averaged over a lager period of time such as a week, a month, multiple week, multiple months, etc. , 5) using a population separate from the population of interest to evaluate those features and determine to what extent various features can be used to represent the behavioral characteristics of a condition of interest, 6) selecting combinations of important features to generate an “index”, and 7) using the “index” to measure changes in the condition over time.

[0010] Technologies disclosed herein may be used to create a behavior-based index for evaluating health conditions and / or life stages and the approaches described herein may be applied to other conditions such as mobility, anxiety, gastrointestinal conditions, renal and / or urinary conditions, food allergy conditions, heart conditions, cancer conditions, weight conditions, oral health conditions, neurological conditions, and / or different life stages (e.g., aging conditions, or differenceswithin a given condition as it manifests at different ages such as puppies with dermatological issues, instead of, or in addition to, adult animals with dermatological issues).

[0011] Technologies are disclosed for determining a behavior-based index for evaluating health conditions and / or life stages for a plurality of subjects. One or more processing devices may be configured to perform one or more techniques. One or more techniques may comprise receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time.

[0012] One or more techniques may comprise identifying, as a second data, a motion, a behavior, and / or a movement exhibited by each of the plurality of subjects over the period of time based, at least in part, on the first data.

[0013] One or more techniques may comprise selecting, as a third data, important features from the first data that can be used to characterize health conditions and / or life stages.

[0014] One or more techniques may comprise inputting, into at least one machine-learning (ML) model, the second data and / or the third data. One or more techniques may comprise processing, by the at least one ML model, the second and / or the third data. One or more techniques may comprise determining, via the at least one ML model, changes over time in health conditions and / or life stages for each of the plurality of subjects, based at least in part, on the processing of the second and / or the third data.BRIEF DESCRIPTION OF DRAWINGS

[0015] The elements and other features, advantages and disclosures contained herein, and the manner of attaining them, will become apparent and the present disclosure will be better understood by reference to the following description of various examples of the present disclosure taken in conjunction with the accompanying drawings, wherein:

[0016] FIG. 1 is a is an example chart of one or more analysis techniques discussed in the present disclosure.

[0017] FIG. 2 illustrates an example of days ordered from Monday to Sunday from top to bottom, with the time of the day indicated on X axis and height of spike indicating scratching activity level summed of respective days and normalized on Y axis.

[0018] FIG. 3 is an example flow diagram of at least one technique for determining an effectiveness of a intervention effect of a subject.

[0019] FIG. 4 is a block diagram of a hardware configuration of an example device that may control one or more parts of a pet bed and / or pet health monitoring control device.

[0020] FIG. 5 illustrates an example output of sample derm and “healthy” (e.g., non-derm comparator group) index averaged over time for all dogs in respective groups.

[0021] FIG. 6 illustrates an example time based split model performance on unseen test data.

[0022] FIG. 7 illustrates a bootstrap sampling on patients / subjects and index performance over at least 4 data splits.

[0023] FIG. 8 illustrates an example bootstrap sampling for Index performance over 84 data splits for some or all combinations of selected test derm dogs.

[0024] FIG. 9 illustrates example one or more feature importance from linear model(s).

[0025] FIG. 10 illustrates an example of visually understandable linear and non-linear models.

[0026] FIG. 11 illustrates an example of an intuitive Gradient Boosted Decision Tree (GBDT) symmetric decision tree which explains how a derm classification is made using the input features.

[0027] FIG. 12 illustrates an example train, test, and output sample auc roc score on a Y axis and smooth features window period on an X axis.

[0028] FIG. 13 illustrates an example derm index bootstrap sampling validation using smooth features.

[0029] FIG. 14 illustrates an example derm index application on unseen test dogs.

[0030] FIG. 15 illustrates an example derm index application on unseen test dogs.

[0031] FIG. 16 illustrates an example of visualizing derm index on unseen test dogs using smooth features.

[0032] FIG. 17 illustrates an example Derm index visualization for Subject Study and Healthy SPDS.

[0033] FIG. 18 illustrates an example of a derm index applied to Subject Study and visualizing index output.

[0034] FIG. 19 illustrates an example of a change from baseline, for example by evaluating a derm index on Subject Study .

[0035] FIG. 20 illustrates an example of one-class support vector machine of sample group data.

[0036] FIG. 21 illustrates an example of differentiating a treatment effect using norm excursion.

[0037] FIG. 22 illustrates an example of Norm excursion showing daily improvement (e.g., a treatment effect) in the two sample groups.

[0038] FIG. 23 illustrates an example of a histogram based on the output of a norm excursion analysis.

[0039] FIG. 24 illustrates an example of a treatment effect detected through norm excursion analysis using an index based on optimal features using optimal features.

[0040] FIG. 25 illustrates an example fraction of outliers in treatment period using optimal features.

[0041] FIG. 26 illustrates an example interpreting One-Class-SVM using decision boundaries based on two input features input data and a decision boundary.

[0042] FIG. 27 illustrates a visualizing treatment effect using smooth and optimal features.

[0043] FIG. 28 illustrates an example of a norm excursion using SPDS optimal smooth features.

[0044] FIG. 29 illustrates an example of the quantity of data (number of days) collected for a range of dogs.

[0045] FIG. 30 illustrates example of the degree of completeness of data collected for a range of dogs.

[0046] FIG. 31 illustrates an example of behavior quantification (ex., scratching) for groups of dogs during different times.

[0047] FIG. 32 illustrates an example of the quantity of data (number of days) collected for a range of dogs from SPDS.

[0048] FIG. 33 illustrates an example of the degree of completeness of data collected for a range of dogs from SPDS.

[0049] FIG. 34 illustrates an example of the degree of completeness of data collected for a range of dogs with less than 50 days worth of data from SPDS

[0050] FIG. 35 illustrates examples of the daily median duration of behaviors such as scratching, shaking and walking among dogs with and without dermatitis issues in SPDS.

[0051] FIG. 36A and FIG. 36B illustrates an example of separation of one or more of the individual features that may best distinguish derm dogs from non-derm (healthy ) dogs.

[0052] FIG. 37 illustrates an example of a comparison of behavior patterns with dogs with and without dermatitis.

[0053] FIG. 38 illustrates an example of the interpretation of the results of an norm excursion analysisDETAILED DESCRIPTION

[0054] For the purposes of promoting an understanding of the principles of the present disclosure, reference will now be made to the examples illustrated in the drawings, and specific language willbe used to describe the same. It will nevertheless be understood that no limitation of the scope of this disclosure is thereby intended.

[0055] Changes in health condition and / or life stage may be characterized by changes in one or more clinical features and / or behaviors. Few conditions can be completely assessed using a single characteristic, and often the characteristics of a health condition or life stage may not change at the same rate, in the same direction or at the same time. Therefore, being able to evaluate multiple characteristics of a health condition may be advantageous in obtaining a complete picture of the health condition and having the needed sensitivity to detect response to interventions, such as medications, nutrition interventions, or simply changes in the health condition or life stage over time.

[0056] In one or more scenarios, a lack of a separate population may exist to train models on, in clinical trial analysis. An external population may be used. Through the use of the WEDS population, characterization of healthy and / or challenged individuals may be made perhaps without contaminating the training data with study data. This may leave the classifier / technique free to operate on the study data without concern of over-fitting and / or inherited bias. As used herein, “healthy” does not connote that the subjects are completely without medical conditions, just not the condition under particular study (e.g., non-derm subjects, etc.)

[0057] In one or more scenarios, one or more limitations of regression and / or other average-based methods for capturing variation and / or level of detail in digital health data, like wearables, may exist. One or more techniques described herein (e.g., index -based) may compensate for such limitations and / or lack of a sufficient level of detail.

[0058] In one or more scenarios, “traditional” statistical methods may struggle to account for data with a large number of repeated measures, at a high frequency, and / or which may be prone to noise. One or more techniques described herein may be curated and / or compiled with these challenges that may provide effective approach(es) to analysis of big / large wearables data.

[0059] In one or more scenarios, there may be little available information on what an on what a range of conditions such as aging, dermatitis, mobility issues, anxiety, gastrointestinal conditions, cancer, and other health or life stage conditions may “look like” in terms of quantified behavioral changes as measured using a wearable device wearable device data. One or more techniques described herein may provide for a (e.g., deep) characterization of these phenotypes, perhaps for example before study analysis may being, among other scenarios.

[0060] In one or more scenarios, there may be a lack of (e.g., basic) standardized study design principles for wearables studies / analysis. One or more techniques described herein may implementbaseline period(s), intervention period(s), standardized eligibility criteria, use of minimum data thresholds (e.g., qualifying days, etc.).

[0061] In one or more scenarios, a series of ML models may be used. The first one (or multiple) apply behavior labels to the accel data based on sub-second features of the accel data, or potential short increments of time that are greater than 1 second, maybe a few seconds. In at least some cases this model (or models) may labeling raw accel data. The second ML model works on the behavior labels instead of the raw accel data. This model identifies features in the behavior data that are characteristic of a condition and uses those to form an index. The third ML model evaluates changes in behavior patterns using the index from the second ML model. This third ML model compares the index value before / after a food intervention or compares two different groups of pets.

[0062] Stated somewhat differently, laddering multiple levels of artificial intelligence may be done in the form of ML models, one or more, or each, of which operates on a different layer of the data (e.g., first converting raw accel data to behaviors, then aggregating behaviors to form an index and finally using the index to make comparisons and inferences about differences).

[0063] FIG. 1 is a is an example chart of one or more analysis techniques discussed in the present disclosure.

[0064] Dermatological and other canine health conditions can be monitored using accelerometers. In some cases, simple descriptive statistics may not be sufficient to fully capture differences between groups of dogs or different times of day. For example, the mean or average value over a period of time may obscure a significant change in the daily duration of a behavior over the same time period. In addition, considering a dog with high scratching activity in the morning and low scratching in the night, and another dog with low scratching in the morning and high scratching in the night with the same magnitude of scratching, their average activity may be similar and / or significant timely activity information may be missed.

[0065] Wearable device (“wearables”) activity data (e.g., high frequency ~ 100Hz accelerometer and gyroscope time series data) is a rich and powerful way of observing changes in health and behavior of patients / subjects. Exploiting the richness of the data may use specialized data analysis methodologies. For example, when using such data to make inferences about treatment effects in clinical trials (e.g., observational and / or randomi zed-control led trials), methodologies that are capable of handling the high-dimensionality of the data can be used in order to make inferences about treatment effects (e.g., typically expressed as simpler claims often scalar in nature). These approaches can also be used for studies other than clinical studies, or simply to observe changes inhealth and / or behavior over time in the absence of a specific intervention.

[0066] Described herein are the development and exploration of a number of methods to analyze such data and make inferences. One motivation is to develop and validate a “toolbox” of methods that can be reused to structure the analysis of studies with similar data sources and objectives.

[0067] Described herein is the use of wearables data (e.g., accelerometer data, cameras, beds, feeders, location data, etc.) and activity patterns in subjects (e.g., pets, dogs, cats), to generate a behavior-based index for evaluating health conditions and / or life stages. For example, in a clinical trial, subjects with dermatological conditions and a history consistent with environmental allergies and with a basal amount of scratching may be enrolled. Dogs may be randomly assigned to test (e.g., Group I) and control (e.g., Group II). After enrollment, all dogs may continue to receive their current food for the baseline period and special dog food (e.g., food that may help manage signs of dermatitis such as scratching) may be provided to the test group (Group I) to observe treatment effect while control group (Group II) may continue to receive their current baseline food. If there is a treatment effect, Group I should have an improvement in signs of the dermatological condition from baseline to a intervention effect period, relative to Group II.

[0068] Data features may be collected on dogs in a (e g., totally) separate population (e.g., a Separate Population Dog Study (SPDS) data set, collected in a population separate from the subjects on which application of the determined features and / or indices was made) in order to identify unique and / or new motion and / or behavior-based features of dermatological conditions. These features may then be used to generate an index, which may be better able to describe the state of the condition vs. individual features alone and simpler to interpret than multiple features that have not been combined. The index may then be applied in a clinical trial(s), by comparing index scores collected in the baseline period (e.g., before the dogs were assigned to the study intervention) to the index scores collected in the intervention period (e.g., after the dogs were assigned to the study intervention) to evaluate the effect of the intervention. Learnings may be applied to other conditions such as aging, mobility, anxiety, gastrointestinal conditions, renal and / or urinary conditions, food allergy conditions, heart conditions, cancer conditions, weight conditions, oral health conditions, neurological conditions, and / or different life stages (e.g., aging conditions, or differences within a given condition as it manifests at different ages such as puppies with dermatological issues, instead of, or in addition to, adult animals with dermatological issues).

[0069] Previous activity and / or wearables data have been used to study clinical outcomes in animals and humans using machine learning algorithms. Methods such as linear functional analysis,clustering and classification have been shown to be helpful in understanding the outcomes in clinical trials like sleep quality in acute insomnia subjects, circadian pattern in advanced dementia and gait patterns in exercise therapy.

[0070] At least a part of this disclosure considers developing a behavior-based index for evaluating health conditions and / or life stages using descriptive statistical features to classify healthy and / or non-impacted dogs vs. dogs impacted by a particular life stage or health condition (for example, “derm” or “dermatitis behavior indicated”) dogs. At least part of this disclosure, a derm index may be applied to dogs with environmental or food allergies to understand treatment effects different health conditions, and / or life stages. At least a part of this disclosure explores norm excursion using a one-class classifier to understand treatment effects.

[0071] Many health conditions such as pruritic skin diseases are multifactorial and characterized by a range of signs and behaviors, most prominently scratching but also potentially shaking, disrupted sleep, fatigue (e.g., if sleep is very disrupted) and potentially other behavior changes, in addition to signs like irritated, red skin, greasy matted coat, skin excoriations, skin thickening, hair loss, and other characteristics. Therefore, an approach that accounts for multiple characteristics of the life stage and / or health condition may be optimal. With multivariate index analysis, the expression of multiple behaviors can be aggregated into a single meaningful value that is used to identify changes in health conditions, particularly complex health conditions, over time In one example, a derm index may be created using statistical features from wearables data that are determined to be characteristic of the distinction between derm vs healthy dogs.

[0072] In one instance, the multivariate derm index is a Machine Learning (ML) methodology where powerful classification algorithms may be developed to identify subjects that are experiencing a health challenge using descriptive statistical features from the subjects’ daily activity. The multivariate index can be paired with other statistical methods, including the norm excursion approach, to quantify differences between the normal behavior of individual dogs. These approaches used “featurized” versions of the wearables data, based on specific characteristics of the wearables data and not only the total duration of behavior in a single day. For example, although the average total daily scratching time in seconds / day over a period of weeks and / or months might be associated with derm conditions, it may not be the optimal way to fully capture the variation in scratching that differentiates dogs with dermatitis from healthy dogs. Similar patterns may be observed with other characteristics of derm conditions such as shaking, sleeping, or total activity, and similar patterns may be observed with other health or life stage conditions as well.

[0073] In one or more scenarios, FIG. 36A and FIG. 36B illustrate that, among the population and features evaluated , not all data features, even those that may be thought to be strong indicators of derm issues (such as scratching, shaking and sleep / rest data features) are equally strongly associated with the presence of reported derm issues. In addition, even within a given behavior category, such as scratching, some data features appear to be more strongly related to the presence and / or absence of derm issues than other features within the same behavior category. Similar findings were observed for features in the shaking and sleep / rest behavior categories. Relatively simple data features such as the total daily duration of the behavior (ex. total daily scratching time in seconds / day) have been commonly used and often taken to be a good representation of the severity of a particular condition such as derm conditions. However, FIG 36 illustrates that other data features, particularly those related to variability such as interquartile range, maximum, minimum, standard deviation as well as those that relate to behavior intensity or the timing of behavior expression (such as the presence of intense bouts or the expression of a behavior during particular times of the day such as daytime, nighttime, or within specific quartiles or even hours of the day) may be more strongly associated with the presence and / or absence of a behavior than total daily duration of the behavior. Some data features may be associated with the presence of a condition, and some data features may be associated with the absence of a condition. Given the complex and multifactorial nature of conditions like dermatitis it may be that approaches that consider multiple features are better able to capture the characteristics of the condition, and thus better able to track changes in the state of the condition over time. For example, when combinations of features are used, a combination that includes scratching, shaking and sleeping behavior may best identify dogs with dermatitis (e.g., specific features listed herein).

[0074] In one or more scenarios, changes in pet health may be measured by tracking a single behavior, but often health states are complex and characterized by changes in multiple behaviors. Dermatitis, such as dermatitis resulting from allergies to environment or food, is one such complex behavior potentially characterized by changes in scratching, shaking, sleeping, sleep quality, rest, and more. This kind of multifaceted, holistic assessment is similar to the way a pet parent or veterinarian might evaluate an animal. In order to accurately measure changes in dermatitis similar to how a pet parent or veterinarian would evaluate a pet, a multivariate index that is capable of measuring changes in multiple behaviors is needed. The derm index has been created using machine learning to identify the pet wearable features that best distinguish healthy animal s / subjects from those with dermatitis.

[0075] To develop an index that represents the severity of pruritic behavior expression, data from the separate population was used to identify behaviors that were increased, decreased and unchanged in dermatitis A gradient boosted decision tree was trained on the data to separate dogs into groups labelled ‘derm’ and ‘healthy ’ according to the output of the multivariate index. Several robustness checks were performed on the results of these experiments. One of these checks was the deliberate mislabeling of healthy dogs as derm dogs before repeating attempts to separate these groups. Where the classifier was found not to be robust for mislabeling, the approach employed for splitting the test and train datasets may be adjusted, with the expectation that curating the data that the model was trained on might provide a more robust model of pruritic behavior, perhaps while still excluding data from dogs that the model might not (e.g., would never) be exposed to other than in validation. A bootstrap sampling robustness check may be performed by training the index on all possible combinations of derm dogs (e.g., a split-by-animal approach).

[0076] To further confirm the ability of the gradient boosted decision tree to separate the groups, a t-test on the sampling means of healthy and derm groups may be performed to test the null hypothesis that independent samples from both groups may / would have identical mean values. To further refine the ability of the index to distinguish between healthy and derm dogs, used smoothed features (e.g., averaged over a window period of days) may be used. A range of smoothing windows may be used to identify the most suitable. To place the index in context and demonstrate its applicability to clinical data, among other reasons, the index for all dogs enrolled in the clinical trial may be calculated. One or more, or each, of these dogs had known skin conditions and / or may / should be distinguishable with the derm index.

[0077] At least another usefulness / objective of the present disclosure may be a set of robust generalizable methodologies which can form a “toolkit” for the analysis of the canine (e.g., among other subjects, other animals, humans, etc.) health data.

[0078] A multivariate derm index is a Machine Learning (ML) methodology where classification algorithms may be developed to distinguish between healthy and derm using descriptive statistical features from daily activity of dogs. Such methods can be applied to the studies of canine health that use accelerometer data to understand treatment effects.

[0079] Norm Excursion is a Machine Learning (ML) methodology where treatment effects may be found by training machine learning classified s) per patient, perhaps on (e.g., only one or at least one) class (e.g., baseline period) and / or outliers may be observed in treatment period (e.g., outliers may identify the treatment effect from a baseline). This methodology may be used, perhaps along withwearables data, to understand gait patterns in exercise therapy which show / may have showed promising results in clinical trial outcomes. This methodology could be used, along with wearables data, to understand changes in a range of companion animal health conditions

[0080] To apply Functional Linear Modelling, high-resolution (One-Second level) scratching activity data may be extracted from interval-based activity labelling for patients in Treatment vs Control. One or more, or each, One-Second Scratching timestamp is labelled (Baseline Control, Study Control, Baseline Treatment , Study Treatment ) based on baseline / study dates. (See Table 1).Table 1 : One-Second scratching activity dog data labelled using group and time period.

[0081] To observe treatment effect, 7 week days (e.g., from Monday to Sunday) of scratching activity data were averaged into a single 24 hour profile, which produced a single scratching activity pattern for each patient from baseline to intervention effect period which is normalized for comparison. FIG. 2 illustrates an example of days ordered from Monday to Sunday from top to bottom, with the time of the day indicated on X axis and height of spike indicating scratching activity level summed of respective days and normalized on Y axis. Plots on the left side of FIG.2 represent baseline period(s). Plots on the right side may represent intervention period(s).

[0082] In one or more scenarios, for example to observe scratching activity patterns over a 24 hour profile, scratching activity at each minute of the day (e.g., Summed over all days) was considered to get a (e.g., single) activity pattern for one or more, or each, subject. To observe a smooth scratching activity pattern, a polynomial regression (Curve fitting) was used on the minute of the day aggregation over each subject. This suggests that scratching starts to increase in the afternoon for asubject and / or peaks during night time. In one or more scenarios, from plots such as those shown in FIG. 2, a detailed understanding of the timing and sequence (e.g., pattern) of a range of animal behaviors that may be characteristic of a health or life stage condition can be observed. Individual deviations from typical timing and sequence (e.g., , patterns) of behavior characteristic of a given life stage or health condition may be observed

[0083] Using a linear functional analysis, smooth patterns (e.g., scratching activity pattern) may be observed and such timely differences can be studied. Using minute of the day scratching summed over all days, the smooth activity pattern (e.g., Scratching) can be studied for a subject in baseline vs. intervention effect time period.

[0084] To observe a smooth (e.g., Linear functional model) scratching pattern for each group by minute of the day (e.g., Summed over all days), a smooth activity pattern for each subject per group may be averaged to produce a single pattern which would describe each group. One or more, or each group may be based on the Control / test subject and time period of the study. It might be observed that scratching reduces during the nighttime and peaks during the evening. However, scratching activity over a 24 hour period may remain similar for baseline intervention effect vs study_intervention effect which may suggest no treatment effect. Ihe index approach with functional data analysis / linear functional modeling may be sued and model the changes in a health index (e.g., the derm index) over the course of a time increment such as a day, week, hour, month etc. The index approach with the responder analysis, the index approach with functional data analysis / functional linear modeling, the index approach with norm excursion (e.g., one or more in combination).

[0085] One or multivariate derm index may have application to analyses of health conditions in dogs, such as dermatitis. One or more classification algorithms may be supervised learning models used to find a useful (e.g., the best) separation between at least two known classes. Given activity patterns for healthy dogs and those with dermatitis issues, machine learning classifiers can find the best fit which separates the two classes (e.g., Healthy and Derm). SPDS data which represents the differences in activities between healthy and derm classes can be used to create a derm index which may give a probability score whether a dog belongs to a healthy or derm group. SPDS data comes as, for example, a daily wearables activity data for dogs who belong to a healthy or derm group, which includes descriptive statistical features over a day for each dog, etc.

[0086] At least some of these features are daily activity based descriptive statistical features computed from accelerometer data and some relate to sleeping, scratching and / or shaking. Table 2illustrates some example data / parameters.Table 2: Examples of wearables data that may be used to construct a Derm Index

[0087] One or more of these descriptive statistical features may be calculated on a daily basis using activity data from accelerometer sensors. These descriptive statistical features may be useful in creating a robust derm index. One or more of these features may provide (e.g., simple and / or detailed) summaries for various activities performed by a dog in a day, like resting, sleeping, walking, running, scratching and / or shaking etc. Referring now to FIG. 3, a diagram 300 illustrates an example technique for determining the effectiveness of a intervention effect , or that the intervention effect is relatively ineffective, for each of a plurality of subjects. The method may be performed by a pet bed and / or pet health monitoring control device, among other devices. The pet bed control and / or pet health monitoring control device may be in communication with at least one camera device and / or a display device. At 302, the process may start or restart.

[0088] At 304, the pet bed control device may receive a first data from a plurality of activity monitors respectively associated with a plurality of subjects. Some, or each, of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. At 306, the pet bed control device may, for each of the plurality of subjects, receive a second data identifying one or more indicators of one or more of a life stage or a life-quality condition for which each of the plurality of subjects received the at least one of a therapy, a nutrition, or a intervention effect over at least some portion of the period of time.

[0089] At 308 the pet bed control device may input, into at least one machine-learning (ML) model, the first data and the second data. At 310, the pet bed control device may process, by the at least one ML model, the first data and the second data. At 312, the per bed control device may determine, via the at least one ML model, that the intervention effect is relatively effective, or that the intervention effect is relatively ineffective, for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. At 314, the process may stop or restart.

[0090] In one or more scenarios, one or more techniques may comprise identifying, as a seconddata, a motion, a behavior, and / or a movement exhibited by each of the plurality of subjects over the period of time based, at least in part, on the first data.

[0091] One or more techniques may comprise selecting, as a third data, important features from the first data that can be used to characterize health conditions and / or life stages.

[0092] One or more techniques may comprise inputting, into at least one machine-learning (ML) model, the second first data and / or the third second data. One or more techniques may comprise processing, by the at least one ML model, the second and / or the third data first data and the second data. One or more techniques may comprise determining, via the at least one ML model, changes over time in health conditions and / or life stages a skin condition for each of the plurality of subjects, based at least in part, on the processing of the second and / or the third data first data and the second data.

[0093] FIG. 4 is a block diagram of a hardware configuration of an example device that may function as a process control device / logic controller. One or more of the disclosed techniques may be processed and / or controlled by the hardware configuration 400. The hardware configuration 400 may be operable to facilitate delivery of information from an internal server of a device. The hardware configuration 400 can include a processor 410, a memory 420, a storage device 430, and / or an input / output device 440. One or more of the components 410, 420, 430, and 440 can, for example, be interconnected using a system bus 450. The processor 410 can process instructions for execution within the hardware configuration 400. The processor 410 can be a single-threaded processor or the processor 410 can be a multi -threaded processor. The processor 410 can be capable of processing instructions stored in the memory 420 and / or on the storage device 430.

[0094] The memory 420 can store information within the hardware configuration 400. The memory 420 can be a computer-readable medium (CRM), for example, a non-transitory CRM. The memory 420 can be a volatile memory unit, and / or can be a non-volatile memory unit.

[0095] The storage device 430 can be capable of providing mass storage for the hardware configuration 400. The storage device 430 can be a computer-readable medium (CRM), for example, a non-transitory CRM. The storage device 430 can, for example, include a hard disk device, an optical disk device, flash memory and / or some other large capacity storage device. The storage device 430 can be a device external to the hardware configuration 400.

[0096] The input / output device 440 may provide input / output operations for the hardware configuration 400. The input / output device 440 (e.g., a transceiver device) can include one or more of a network interface device (e.g., an Ethernet card), a serial communication device (e.g., an RS-232 port), one or more universal serial bus (USB) interfaces (e.g., a USB 2.0 port) and / or a wireless interface device (e.g., an 802.11 card). The input / output device can include driver devices configured to send communications to, and / or receive communications from one or more networks. The input / output device 400 may be in communication with one or more input / output modules (not shown) that may be proximate to the hardware configuration 400 and / or may be remote from the hardware configuration 400. The one or more output modules may provide input / output functionality in the digital signal form, discrete signal form, TTL form, analog signal form, serial communication protocol, fieldbus protocol communication and / or other open or proprietary communication protocol, and / or the like.

[0097] The camera device 460 may provide digital video input / output capability for the hardware configuration 400. The camera device 460 may communicate with any of the elements of the hardware configuration 400, perhaps for example via system bus 450. The camera device 460 may capture digital images and / or may scan images of various kinds, such as Universal Product Code (UPC) codes and / or Quick Response (QR) codes, for example, among other images as described herein. In one or more scenarios, the camera device 460 may be the same and / or substantially similar to any of the other camera devices described herein.

[0098] The camera device 460 may include at least one microphone device and / or at least one speaker device. The input / output of the camera device 460 may include audio signals / packets / components, perhaps for example separate / separable from, or in some (e.g., separable) combination with, the video signals / packets / components the camera device 460.

[0099] The camera device 460 may also detect the presence of one or more people that may be proximate to the camera device 460 and / or may be in the same general space (e.g., the same room) as the camera device 460. The camera device 460 may gauge a general activity level (e.g., high activity, medium activity, and / or low activity, light level, presence or absence of other people or animals or objects of interest, such as a food bowl or litterbox) of one or more people, subjects, animals, etc., that may be detected by the camera device 460. The camera device 460 may detect one or more general characteristics (e.g., height, body shape, skin color, pulse, heart rate, breathing count, etc.) of the one or more people, subjects, animals, etc., detected by the camera device 460. The camera device 460 may be configured to recognize one or more specific people, for example.

[0100] The camera device 460 may be in wired and / or wireless communication with the hardware configuration 400. In one or more scenarios, the camera device 460 may be external to the hardware configuration 400. In one or more scenarios, the camera device 460 may be internal to the hardwareconfiguration 400.

[0101] The hardware configuration may process and / or control measurements from other technologies including a smart pet bed, video, images, photos, output from computer vision algorithms running on videos, microphone data, pet parent evaluations, medication information, gyroscope data, and / or accelerometer data.

[0102] Class imbalance may be a common scenario which could occur in many real-world applications where the class distribution of data may be (e.g., highly) imbalanced. Traditional costinsensitive machine learning classifiers may perform poorly on these imbalanced datasets. To address this problem, among other scenarios, one or more different techniques may be applied to address class imbalance, like oversampling minority class and / or class weights, etc., for example. Oversampling is a technique where random oversampling of the minority class (e.g., the Subject Study derm class) may be used to balance the machine learning dataset. Class weights is another approach where one or more weights may be assigned to the minority class which may help address class imbalance.

[0103] Different training and test data split techniques are described herein. Bootstrap sampling cross validation techniques may be used to evaluate the derm index due to the perhaps for example where a relatively low number of derm dogs may be available, for example, among other scenarios. In circumstances where the number of animals with the target health and / or life stage condition is lower than desired, bootstrap sampling cross validation techniques can be used

[0104] In one or more scenarios, randomly splitting data techniques may be used. At least one technique to develop a derm index may include creating a machine learning dataset, where each sample / row is the daily wearables activity of healthy / derm dogs. The training and test datasets may be split randomly.

[0105] For example, in one or more scenarios, perhaps as part of a pre-processing and feature selection, one or more of the following may be done. One or more static features may be removed (e.g., as the model can memorize animals). One or more numeric features may be rescaled. A focus may be applied on features related to scratching and / or sleeping specifically. One or more highly correlated features may be removed. Oversampling may be done to the minority class (e.g., derm) to address class imbalance.

[0106] To help reduce class imbalance between majority (e.g., healthy ) and minority (e.g., derm), among other reasons, the minority class may be oversampled. A gradient boosted decision tree model may be trained on the data. Such a trained model may show good performance on the output of asample test set and / or good separation between the two classes when the derm index is applied on the test set. In one or more scenarios, an Out of sample test set auc roc score may be: 0.79, for example.

[0107] A random split model may work well in separating two groups (e.g., healthy and derm) using the derm index. Clear separation may be seen between two groups with higher probability scores for derm (Class 1) and / or lower probability scores for health (Class 0). FIG. 5 illustrates an example output of sample derm and healthy index averaged over time for all dogs in respective groups. This methodology may show good results to create a good derm index. It may be useful to build further confidence by performing robustness tests and trying various kinds of data splits, for example.

[0108] In one or more scenarios, a time series split may be used. A useful (e.g., more useful) validation of derm index may be a training model on historical data points and predicting into the future (e.g., which may intuitively make sense and builds confidence). The training and test data (Out of Sample) may be split by time in one or more scenarios. Even when the model has never seen the future activities of dogs, it may be able to (e.g., clearly) distinguish between healthy and derm using the index. FIG. 6 illustrates an example time based split model performance on unseen test data (e.g., future data). In one or more scenarios, an example out of sample test set auc roc score may be: 0.88, for example.

[0109] To further build confidence in this model and / or the technique, one or more robustness checks may be performed to understand if the results are in fact promising and, perhaps, not a statistical accident.

[0110] In one or more scenarios, train and test data may be split by animal such that the machine learning model might not see the test dogs and / or may be a good approach to evaluate the model(s) on these unseen test dogs. The model may be trained on some dogs (Train) and may be evaluated on unseen test dogs. At least one consideration while working on splitting the data by animal may be the very few number of derm dogs (e.g., 9 derm dogs), which is described herein.

[0111] In one or more scenarios, the data sets may be split by animal. The machine learning dataset may be split by animal where some animals may be taken out from derm and some animals from healthy groups, that the model might never see in the training phase.

[0112] In one or more scenarios, this may be intuitive as the model might never see the out of sample test animals. Evaluating the derm index based on this methodology may be even more robust. With very few number of derm dogs to work with, a bootstrap sampling robustness check may beperformed where the derm index is trained on one or more, or all, possible combinations of selecting derm dogs for the test set(s). One or more models may be trained on one or more, or all, possible combinations of selecting out of sample derm dogs (e.g., 9 Choose 3 e.g., 84 combinations) (e.g., 84 Models). For one or more, or each of these models, healthy and derm group index performance may be average on out (output) of sample test data. This average may then represent performance of individual models (e.g., 84 Models).

[0113] In one or more scenarios, using a bootstrap sampling on patients / subjects and index group mean (e.g., Healthy and Derm), for example, a bootstrap distribution of index performance over data splits may show clear separation between healthy and derm dogs which suggests that the derm index works well in at least one or more scenarios.

[0114] FIG. 7 illustrates a bootstrap sampling (e.g., a robustness check) on patients / subjects and index performance over at least 4 data splits (e.g., some or all combinations of selecting test derm dogs). In one or more scenarios, a significant difference between the two samples (e.g., derm and healthy ) means with P ~ 10A(- 16) (P<0.01). Bootstrap sampling on patients / subjects suggests good derm index when index performance are visualized over bootstrap data splits (see FIG. 7).

[0115] FIG. 8 illustrates an example bootstrap sampling for Index performance for some or all combinations (e.g., 84 data splits) of selected test derm dogs. The derm index may perform well in distinguishing between derm dogs and healthy dogs along multiple data splits (e.g., bootstrap sampling cross validation). From FIG. 8, the model may produce higher probability scores for derm and pushes for lower probability scores for healthy dogs which suggests that using split by animal methodology derm index may work well on test dogs which it never saw.

[0116] In one or more scenarios, some or more features may be used to develop a derm index, such as ['max_allday_sleep', 'intense_scratching_bouts', 'median_ allday _sleep', 'intense_shaking_bouts'], for example.

[0117] One or more t-tests may be performed on this bootstrap sampling means of healthy and derm groups. Using t-test(s) may show a (e.g., significant) amount of difference between means of derm and healthy groups. This may be a t-test(s) for the null hypothesis that two independent samples have identical average values. T-test may reject null hypothesis and suggests at least one alternative hypothesis that there is significant difference between two samples (e.g., derm and healthy ) means with P ~ 10A(-16) (P<0.01). In one or more scenarios, t-test suggests that the derm index may work well in identifying derm and healthy dogs.

[0118] In one or more scenarios, t-test may be performed on this robustness check to validate thesplit by animal methodology / technique. Using t-test(s) might not show a significant amount of difference between means of derm and healthy groups, perhaps because in this robustness check healthy dogs (e.g., only healthy dogs) may be used as the data sample, and / or derm labels and healthy labels may be randomly assigned. A t-test for the null hypothesis may include at least two independent samples that may have identical average values. T-test(s) may suggest against the rejection of the null hypothesis in that there is no significant difference between two groups healthy and derm. There may be no significant difference of mean between both groups with p-value>0.01 (P value- 0.03).

[0119] Such a robustness check suggests that the original results with split by animal methodology may work well when creating a derm index. This derm index may show promising results in distinguishing derm dogs and healthy dogs, for example in general using daily wearables activity data.

[0120] One or more linear models may be (e.g., relatively simple) interpretable models. One or more logistic regression model(s) may distinguish between healthy dogs and derm dogs. Feature scaling, removing correlated features, and feature selection may be done as a preprocessing. This may produce descent and consistent auc roc along train, test and output of sample test sets. Results also indicate intense scratching bouts, night time scratching, and moderate scratching bouts push for derm dogs.

[0121] FIG. 9 illustrates example one or more feature importance from linear model(s) (e.g., logistic regression) (e.g., derm index).

[0122] Linearly separable data is data that can be classified into different classes by drawing a (e.g., simple) line (or a hyperplane) through the data. But sometimes the data cannot be separated just by drawing a simple line, as there may be a non-linear relationship between the input features. In one or more scenarios, linear models may be too simple to be able to adequately model real world systems. This is shown in the example of FIG. 10. FIG. 10 illustrates an example of visually understandable linear and non-linear models.

[0123] In one or more scenarios, a derm index that may work well (e.g., best) using the gradient- boosted decision trees (GDBT) model (e.g., a nonlinear model). GDBT is a boosting technique which uses gradient boosting which is a useful (e.g., powerful) way of producing an accurate prediction rule in the form of an ensemble of weak prediction models (e.g., decision trees). An overfitting detector can be used to stop the training earlier than the training parameters dictate. This may help to create a good derm index where data has non-linear separation. FIG. 11 illustrates an example ofan intuitive GDBT symmetric decision tree which explains how a derm classification is made using the input features.

[0124] In one or more scenarios, smooth features, such as features averaged over a window of a period of time (e.g., days, hours, weeks, months, etc.) were considered to perhaps produce a better model(s).

[0125] Smooth features may be calculated as a rolling mean over n number of days / weeks / hours / months etc., which in one or more scenarios may be averaging days worth of data and using that as the input feature. Using the smooth features may be useful as there may be less spikes in the input features which may provide the algorithm some capacity (“breathing room”) to fit the input data. The algorithm might not be under pressure to fit otherwise extreme spikes in the input data. Using smooth features may be a useful strategy for developing a good / effective derm index.

[0126] In one or more scenarios, using bootstrap sampling 18% of the time there is an increase in test out of sample AUC when using smooth features, for example when checked with some, or all, possible combinations of selecting derm dogs (n=84).

[0127] Smooth features may be quite stable when used on the derm index, the difference of mean histogram (e.g., using bootstrap sampling) may be (e.g., always) away from zero (e.g., positive). This suggests that the model developed using smooth features may be able to correctly distinguish between healthy dogs and derm dogs. Using smooth features, AUC increases linearly with the window period (e.g., the number of days a feature is averaged) until a point and then decreases. This suggests smooth features may improve the performance of the model. FIG. 12 illustrates an example train, test and output sample auc roc score on a Y axis and smooth features window period on an X axis.

[0128] T-test was performed on this bootstrap sampling means of healthy and derm groups. Using t-test may show a (e.g., significant) amount of difference between means of derm and healthy groups. In one or more scenarios, this is a t-test for the null hypothesis that two independent samples may have identical average values.

[0129] T-test(s) may reject the null hypothesis and may suggest an alternative hypothesis that there is a significant difference between two samples (derm and healthy ) means with P ~ 10A(-108) (P<0.01). In one or more scenarios, t-test(s) may suggest that the derm index works well in identifying derm and healthy dogs. In one or more scenarios, results may be even better using more data from the latest SPDS extract.

[0130] In one or more scenarios, a derm index may be formed using smooth features on a latestextract. As more data may show promising derm index, smoothing the input features and / or developing a derm index on top of smooth features may be performed. Smooth features may be calculated as a rolling mean over n number of days, which may be averaging days worth of data and using that as the input feature. Such techniques using the smooth features may be useful as there may be less spikes in the input features which would give the algorithm some capacity (“breathing room”) to fit the input data. The algorithm might not be “under pressure” to fit these extreme spikes in the input data. Using smooth features may be a useful strategy for developing a good / effective derm index.

[0131] Using smooth features along with bootstrap on patients / subject, an index may (e.g., clearly) distinguish between derm dogs and healthy dogs which the model has never seen / processed before. This clear separation between two groups using smooth features may be useful, as shown in FIG. 13. FIG. 13 illustrates an example derm index bootstrap sampling validation using smooth features (e.g., on latest SPDS extract). A significant difference may occur between two samples (Derm and healthy ) means with P ~ 10A(-203) (P<0.01).

[0132] T-test(s) may reject null hypothesis and may suggest an alternative hypothesis that there is significant difference between two samples (Derm and healthy ) means with P ~ 10A(-203) (P<0.01). T-test(s) may suggest that the derm index works well in identifying derm and healthy dogs. Results may be better using more data from the latest SPDS extract along with smooth features.

[0133] One or more scenarios may employ visualizing the application of derm index (e.g., using smooth features) on unseen test dogs over time. As shown in FIG. 14, the derm index on an unseen test dog (e.g., healthy ) shows the derm index may be able to identify healthy dogs by pushing the probability of derm lower. FIG. 14 illustrates an example derm index application on unseen test dogs (e.g., performance) (e.g., healthy ).

[0134] FIG. 15 illustrates an example derm index application on unseen test dogs (e.g., performance) (e.g., derm). As shown in FIG. 15 illustrates, the derm index may be applied on an unseen test dog (e.g., derm). As shown, the derm index is able to classify this dog as derm by pushing the probability of derm higher. FIG. 16 illustrates an example of visualizing derm index on unseen test dogs using smooth features / smoothening (e.g., rolling mean window: 7).

[0135] In one or more scenarios, a multivariate derm index may be used on Subject Study data. This multivariate derm index may be applied to subject study data to understand dogs under study from baseline to a intervention effect period. Initially all dogs (e.g., purple and green) may be picked based on CADLI (Canine Atopic Dermatitis Lesion Index) and / or PVAS (Pruritus Visual AnalogScore) to enroll in the study, so they have basal scratching. At least some, or all, of these dogs have known skin conditions so they may be distinguishable with a derm index which is the case shown in FIG. 17. FIG. 17 illustrates an example Derm index visualization for Subject Study and Healthy SPDS. A Significant difference may be seen between means of All Subject Study (Green and Purple) and outSample Healthy mean (SPDS) with P-value<0.01 (P -value- 10A(-l 6)).

[0136] In one or more scenarios, such as for the intervention effect in the purple group, a intervention effect period of purple group may be (e.g., should) be near zero (e.g., healthy ) and a baseline period towards one (e.g., derm). But that might not be the case as shown in FIG. 18. FIG. 18 illustrates an example of a derm index applied to Subject Study and visualizing index output. No significant difference of mean may be seen between baseline and intervention effect purple with P- value>0.01 (P -value = 0.5869).

[0137] In one or more scenarios, a multivariate derm index was developed using these features from daily activity from wearables:['max allday sleep', 'intense scratching bouts', 'median allday sleep', 'intense shaking bouts']

[0138] In one or more scenarios, checking intervention effect may use a derm index on Subject Study (e.g., change from baseline). In such scenarios, SPDS derm index may be used to classify Subject Study patients as derm or healthy . For each patient / subject, the difference may be computed in an average index between a study period and a baseline period (e.g., change from baseline mean (CFB)).

[0139] FIG. 19 illustrates an example of a change from baseline, for example by evaluating a derm index on Subject Study . No significant difference may be seen between two groups using CFB with p value> 0.01 (P-value=0.9668). When the difference in mean of CFB between purple and green is tested, if there is a intervention effect the purple group may (e.g., should) be more towards negative (e.g., decreased derm). In one or more scenarios, a positive may indicate increased derm behavior and negative may indicate a intervention effect . In one or more scenarios, there might be no significant separation between the two groups purple and green. Using t-test, there may be no significant difference between two groups using CFB with p value greater than 0.01 (P- value=0.9668), for example.

[0140] One or more scenarios consider norm excursions. Regarding the Subject Study , during the baseline period dogs belonging to both groups (e.g., Purple / Group I and Green / Group II) may havehigh scratching behavior as these dogs are selected using standard procedures like CADLI (Canine Atopic Dermatitis Lesion Index) and / or owner reported PVAS (Pruritus Visual Analog Score). For example, if there is a intervention effect in the purple group / group I , among other scenarios, then baseline purple vs intervention effect purple may be (e.g., easily) distinguished in terms of scratching patterns and / or other activity patterns. Using One-Class-SVM classifier makes may be useful, as training the model on (e.g., just on) the baseline period and / or looking for outliers (e.g., a intervention effect) in the intervention effect period may explain / understanding clinical outcome(s) (e.g., treatment effect(s)).

[0141] FIG. 20 illustrates an example of one-class classification of sample group data. In One- Class-SVM, the classifier is (e.g., only) trained on the negative class (e.g., in a Subject Study baseline period) that may be taken as “normal” or “in-liners”, for example. It may be inferred how many outliers are present in the intervention effect period to understand if there is a treatment effect. For example, if there is a treatment effect, dogs in the purple group I (e.g., which are given special antiscratch food) may have more outliers than dogs in the green group II in the intervention effect period. SVM may be used to separate two classes using a hyperplane with the largest possible margin. One- Class-SVM may use a hypersphere to encompass all the “normal” or “inliner” instances. One or more, or any, instance lying outside the hypersphere may be outliers.

[0142] To study the treatment effect, One-Class-SVM model(s) were trained on the baseline period and / or observed outliers in the intervention effect period. Using such techniques, more positive treatment effects may be observed in purple group I over time than in green group II which suggests improvement in pruritus behavior when dogs are given special dog food (e.g., purple group I).

[0143] In one or more scenarios, treatment effects using SPDS index features may be considered. FIG. 21 illustrates an example of understanding treatment effect using norm excursion (e.g., proportion of outliers in a intervention effect period). Purple group I confidence interval is (e.g., based on fraction of outlier) (0.5210, 0.5877). Green group II confidence interval is (e.g., based on fraction of outlier) (0.4321, 0.4964).

[0144] One-Class-SVM using daily activity features (SPDS features) models may include one or more of:['max_allday_sleep', 'intense_scratching_bouts', 'median_allday_sleep','intense shaking bouts']

[0145] The features that were used to train the One-Class-SVM are consistent with features usedto build the derm index to distinguish between healthy dogs and derm dogs, which may make results from derm index consistent with norm excursion.

[0146] Norm excursion may show decent improvement in a treatment effect in purple group I when observed on a daily level, for example, as shown in FIG. 22. FIG. 22 illustrates an example of Norm excursion showing daily improvement (e.g., treatment effect) in the purple group I and the green group II.

[0147] Using the SPDS features, FIG. 23 illustrates a histogram where each point is the fraction of outliers in the intervention effect period which may suggest there is a treatment effect. There seems to be a good separation for proportion of outliers between two groups (Purple / Group I and Green Group II). FIG. 23 illustrates an example of a proportion of outliers in a intervention effect period using a norm excursion histogram. No significant difference in proportion of outliers may be seen between two groups (e.g., Purple / Group I and Green / Group II) with p value greater than 0.05 (P- value=0.1072).

[0148] T-test(s) suggest that there is no significant difference in proportion of outliers between two groups (e.g., Purple / Group I and Green / Group II) with p value greater than 0.05 (P-value=0.1072). The null hypothesis might not be rejected in that there might be no significant difference between the proportion of outliers between two groups (e.g., Purple / Group I and Green / Group II).

[0149] In one or more scenarios, norm excursion made be done using one or more useful (e.g., optimal) features. Using optimal features which may be, for example, median daytime scratching and / or intense scratching bouts, a more significant treatment effect in the purple group I may be observed in FIG. 24. FIG. 24 illustrates an example of a treatment effect using optimal features. A One-Class-SVM model may be employed using daily activity features (e.g., optimal features) such as [median_daytime_scratching, intense_scratching_bouts], etc., for example.

[0150] Using the intuition that there is more area under purple line than in green line suggests there is significant treatment effect in purple group I. A “brute force” type of search may be made to find the optimal features which had significant separation between them. FIG. 23 illustrates an example plot for some or most optimal features.

[0151] FIG. 25 illustrates a histogram where one or more, or each, point is the fraction of outliers in the intervention effect period which suggests there is a treatment effect. For one or more, or most, optimal features, there seems to be a good separation for proportion of outliers between two groups (e.g., Purple / Group I and Green / Group II). FIG. 25 illustrates an example fraction of outliers in intervention effect period using optimal features (e.g., a norm excursion).

[0152] Using the t test to check if there is significant difference between two groups (e.g., proportion of outliers), there is good evidence that they differ significantly with p value less than 0.05 (P-value=0.03088).

[0153] In one or more scenarios, a binomial test (e.g., difference in proportion) may be performed. To increase confidence in norm excursion, among other reasons, a binomial test may be used for a difference in proportion at significance level 0.05, for example. The t-test performed previously effectively weights each patient equally. Some patients / subjects have fewer days in the study period, but a t-test would weigh these patients equally. All patient-days may be pooled producing a binary Bernoulli outcome for each patient / subject-day. Such techniques may weigh each patient / subj ectday equally. A test for difference in proportions can measure the treatment effect. Such techniques suggest even more statistically significant treatment effects.

[0154] Described below are at least some results for an optimal features model:Feature: ['median_daytime_scratching', 'intense_scratching_bouts'] HO: P1=P2 (pl is same as p2)Ha: Pl !=P2 (pl is not same as p2) Results:P-value: < 0.0001

[0155] The null hypothesis may be rejected if pl is the same as p2. In one or more scenarios, there may be a significant difference.

[0156] Described below are the results for SPDS index features:Features: ['max_allday_sleep', 'intense_scratching_bouts', 'median_allday_sleep', 'intense shaking bouts']HO: P1=P2 (pl is same as p2)Ha: Pl !=P2 (pl is not same as p2) Results:P-value:< 0.0001

[0157] The null hypothesis may be rejected if pl is the same as p2. In one or more scenarios, there may be a significant difference.

[0158] A binomial test (difference in proportion test) suggests that using optimal features and / or SPDS derm index features in a norm excursion shows a significant treatment effect in test subjects (e.g., purple / Group I) than in green / Group II (e.g., control subjects).

[0159] In one or more scenarios, interpreting one-class-SVM linear model(s) on optimal features may be performed. To verify that less scratching may correspond to a treatment effect and not the other way around, the SVM model’s linear decision boundary may be visualized using what may classify as whether a data point belongs to in-liner (e.g., regular observation) and / or outliers (e.g., abnormal observation and / or treatment effect or reduced scratching), for example:Linear one-class-svm Equation: 16*xl+30.5*x2, where xl = median_daytime_scratching x2= intense_scratching_bouts

[0160] It may be (e.g., intuitively) clear that outlier detection in norm excursion may correspond to less scratching as shown in FIG. 26. FIG. 26 illustrates an example interpreting One-Class-SVM using input data and a decision boundary.

[0161] In FIG. 26, the red line may be the learned decision boundary which separates in-liners (e.g., a baseline period) from outliers (e.g., yellow) (e.g., a intervention effect period). Outliers detected by the one class SVM model are close to zero. This may (e.g., intuitively) make sense that less scratching behavior in the intervention effect period corresponds to significant treatment effect (e.g., less scratching) from the baseline period.

[0162] In one or more scenarios, norm excursion using smooth features may be conducted. In one or more scenarios, norm excursion using smooth features might not show a promising treatment effect. Smooth features may be a rolling means of daily activity features for patients / subjects. Smooth features may be used to study treatment effects using norm excursion. FIG. 27 shows a use of optimal smooth features for the weekly treatment effect for purple / group I vs green / group II.

[0163] FIG. 27 illustrates a visualizing treatment effect using smooth and optimal features, such as ['median daytime scratching', 'intense scratching bouts'], for example (e.g., norm excursion). T- test suggests no significant difference between two groups with p value greater than 0.01 (P -value: 0.4601). In one or more scenarios, this suggests no treatment effect using smooth features on norm excursion.

[0164] In one or more scenarios, SPDS index optimal smooth features might not show a treatment effect as shown in FIG. 28. FIG. 28 illustrates an example of a norm excursion using SPDS optimal smooth features, such as ['max_allday_sleep', 'intense_scratching_bouts','median_allday_sleep','intense_shaking_bouts'], for example. T-test suggests no significant difference between the two groups with p value greater than 0.01 (P-value:0.1560).

[0165] As described herein, one or more machine-language (ML) approaches / techniques to derm index construction may be successfully applied and / or validated for an output of a sample(s). A ML based norm excursion approach employing one class SVMs may be successfully applied to Subject Study data indicating a treatment effect of a intervention effect (e.g., diet, medication, etc ). In one or more scenarios, the LFA analysis may be inconclusive, perhaps because there might not be a differential intraday pattern response discernable in the treatment effect, among other reasons.

[0166] At least three methodologies / techniques described herein may be (e.g., widely) applicable and / or may form the basis for a “toolkit” for the analysis of similar clinical trials. They may represent a foundation upon which a robust set of tools may be built upon and / or added to as may be useful (e.g., necessary).

[0167] As described herein, the results suggest that Machine Learning approach(es) may be an effective lens with which to view the rich, high dimensional data produced by wearable sensors. The ML approach(es) have been demonstrated, for example through index construction analysis, to uncover interpretable concepts among the multitude of features, perhaps automatically, for example. The ML analysis has been demonstrated to be adaptable to formulating and testing specific hypotheses related to clinical trial treatment effects. Those skilled in the art will appreciate that the subject matter described herein may make accessible insights from wearable sensor data that may have been left inaccessible with the methods of more classical statistical analysis of clinical trial data.

[0168] FIG. 29 illustrates an example of the quantity of data (number of days) collected for a range of dogs.

[0169] FIG. 30 illustrates example of the degree of completeness of data collected for a range of dogs.

[0170] FIG. 31 illustrates an example of behavior quantification (ex., scratching) for groups of dogs during different times.

[0171] FIG. 32 illustrates an example of the quantity of data (number of days) collected for a range of dogs from SPDS.

[0172] FIG. 33 illustrates an example of the degree of completeness of data collected for a range of dogs from SPDS.

[0173] FIG. 34 illustrates an example of the degree of completeness of data collected for a range of dogs with less than 50 days worth of data from SPDS

[0174] FIG. 35 illustrates examples of the daily median duration of behaviors such as scratching, shaking and walking among dogs with and without dermatitis issues in SPDS.

[0175] FIG. 36A and FIG. 36B illustrate an example of separation of one or more of the individual features that may best distinguish derm dogs from non-derm (healthy ) dogs.

[0176] FIG. 37 illustrates an example of a comparison of behavior patterns with dogs with and without dermatitis.

[0177] FIG. 38 illustrates an example of the interpretation of the results of an norm excursion analysis.

[0178] In view of FIG. 1 to FIG. 38, in one or more scenarios, one or more computer-implemented methods / techniques / devices / systems / processes of evaluating a intervention effect and / or a change in a condition for a plurality of subjects may be implemented. One or more methods may comprise receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors measuring at least some activity may be conducted by the respective plurality of subjects over a period of time. One or more methods may comprise, for each of the plurality of subjects, receiving a second data identifying one or more indicators of at least one of: a life stage or a life-quality condition for which each of the plurality of subjects received the at least one of: the therapy, the nutrition, or the intervention effect over at least some portion of the period of time. One or more methods may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more methods may comprise processing, by the at least one ML model, the first data and the second data. One or more methods may comprise determining, via the at least one ML model, at least one of: the intervention effect is relatively effective, or that the intervention effect is relatively ineffective, for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. In one or more scenarios, one or more of the aforementioned elements may be performed by one or more processors of one or more computing devices.

[0179] In one or more scenarios, the life-quality condition may be at least one of: anxiety, limited mobility, a gastro-intestinal condition, an allergic condition, a dermatologic condition, a food sensitivity, a liver condition, a weight condition, a heart condition, a joint condition, a thyroid condition, a urinary condition, a glucose management issue, and / or cancer.

[0180] In one or more scenarios, the life stage may be at least one of: an adolescent stage, a reproductive stage, a mid-life stage, and / or a geriatric stage.

[0181] In one or more scenarios, the plurality of subjects may comprise one or more of: a human, or an animal.

[0182] In one or more scenarios, the intervention effect may be at least one of: a therapy, a nutrition, a medical intervention effect, and / or a physical intervention effect.

[0183] In one or more scenarios, the intervention effect may be at least one of: a placebo regimen, a palliative regimen, a wellness regime, and / or a therapeutic regimen.

[0184] In one or more scenarios, a dermatologic condition may comprise at least atopic dermatitis.

[0185] In one or more scenarios, the activity monitor may be at least one of: an animal-body- mounted device, an accelerometer, a human-body-mounted sensor, an animal-wearable sensor, or a human-wearable sensor.

[0186] In one or more scenarios, the at least some activity may comprise at least one of: a behavior of the subject, a motion of the subject, audio data from the subject, heart rate date from the subject, temperature data of the subject, and / or breathing rate or a movement of the subject.

[0187] In one or more scenarios, the at least some activity may comprise at least one of: licking, walking, running, trotting, cantering, galloping, pacing, ambling, limping, abnormal gait, vomiting, urinating, defecating, passing a hairball, straining to urinate, straining to defecate, digging, eating, drinking, chewing, grooming, playing, dressing, bathing, driving, exercising, cooking, scratching, sleeping, sleeping, resting, and / or shaking.

[0188] In one or more scenarios, the plurality of activity monitors may include at least one location sensor, wherein the at least one location sensor may provide at least some location data corresponding to the at least some activity.

[0189] In one or more scenarios, the nutrition may comprise at least one of: a nutrition regimen not specifically formulated for the at least one of: the life stage or the life-quality condition, a nutrition regimen formulated for the at least one of: the life stage or the life-quality condition, and / or a nutrition regimen therapeutically formulated for the at least one of: the life stage or the life-quality condition.

[0190] In one or more scenarios, the at least one ML model may comprise at least a full responder model. The full responder model may be configured to treat time as a categorical variable; and / or to assume a distinct treatment effect at each time point.

[0191] In one or more scenarios, the full responder model may comprise at least eighteen parameters.

[0192] In one or more scenarios, the at least one ML model may comprise at least a reduced responder model. The full responder model may be configured to treat time as a categorical variable; and / or to assume constant treatment effect at each time point.

[0193] In one or more scenarios, the reduced responder model may comprise at least ten parameters.

[0194] In one or more scenarios, the at least one ML model may comprise at least a support vector machine (SVM) classifier model.

[0195] In one or more scenarios, the SVM classifier model is a one-class SVM classifier model.

[0196] In one or more scenarios, one or more methods / techniques / devices / systems / processes may comprise receiving a first assessment of the at least one of: each of the subjects’ life stage, or each of the subjects’ life-quality condition as a third data. The first assessment may be provided by one or more evaluators associated with each of the respective plurality of subjects over the at least some portion of the period of time. The determining, via the at least one ML model, at least one of the intervention effect is relatively effective, or that the intervention effect is relatively ineffective, for each of the plurality of subjects, may be based at least further in part, on the processing of the first data, the second data and the third data.

[0197] In one or more scenarios, the one or more evaluators may be associated with the each of the respective plurality of subjects comprises: a medical practitioner, a therapist, a pet caretaker, a veterinarian, and / or a computer vision algorithm.

[0198] In one or more scenarios, the first assessment of the at least one of: each of the subjects’ life stage, or each of the subjects’ life-quality condition may comprise, at least in part, an assessment provided via one or more assessment tools.

[0199] One or more computer-implemented methods / techniques / devices / systems / processes of determining a skin condition index for a plurality of subjects may be implemented. One or more methods may comprise receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. One or more methods may comprise identifying, as a second data, at least one of: a motion, a behavior, or a movement exhibited by each of the plurality of subjects over the period of time based, at least in part, on the first data. One or more methods may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more methods may comprise processing, by the at least one ML model, the first data and the second data. One or more methods may comprisedetermining, via the at least one ML model, a skin condition for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. In one or more scenarios, one or more of the aforementioned elements may be performed by one or more processors of one or more computing devices.

[0200] In one or more scenarios, the skin condition may be at least one of: an allergic condition, or a dermatological condition.

[0201] In one or more scenarios, the skin condition may be atopic dermatitis.

[0202] In one or more scenarios, the skin condition may be at least one of: an animal skin condition, or a human skin condition.

[0203] In one or more scenarios, the activity monitor is at least one of: an animal-body-mounted device, an accelerometer, a human-body-mounted sensor, an animal-wearable sensor, or a humanwearable sensor.

[0204] In one or more scenarios, the at least one of: a motion, a behavior, or a movement may comprise at least one of: licking, walking, running, trotting, cantering, galloping, pacing, ambling, limping, abnormal gait, vomiting, urinating, defecating, passing a hairball, straining to urinate, straining to defecate, digging, eating, drinking, chewing, grooming, playing, dressing, bathing, driving, exercising, cooking, scratching, sleeping, sleeping, resting, and / or shaking.

[0205] One or more computer-implemented methods / techniques / devices / systems / processes of of evaluating a skin condition intervention effect for a plurality of subjects may be implemented. One or more methods may comprise receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. One or more methods may comprise, for each of the plurality of subjects, receiving an indicator as a second data identifying a type of skin condition intervention effect that each of the plurality of subjects received over the period of time. One or more methods may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more methods may comprise processing, by the at least one ML model, the first data and the second data. One or more methods may comprise determining, via the at least one ML model, at least one of: that the indicated skin condition intervention effect is relatively effective, or that the indicated skin condition intervention effect is relatively ineffective, for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. In one or more scenarios, one or more of theaforementioned elements may be performed by one or more processors of one or more computing devices.

[0206] In one or more scenarios, the skin condition intervention effect may include at least one of: a nutritional regime, a placebo regime, a shampoo regime, a supplement regime, and / or a medicinal regime.

[0207] In one or more scenarios, the skin condition is at least one of: an allergic condition, or a dermatological condition.

[0208] In one or more scenarios, the skin condition may be atopic dermatitis.

[0209] In one or more scenarios, the skin condition may be at least one of: an animal skin condition, or a human skin condition.

[0210] In one or more scenarios, the activity monitor may be at least one of: an animal collarmounted device, an accelerometer, and / or one or more human-body-mounted sensors.

[0211] In one or more scenarios, the at least some activity may be at least one of: a behavior of the subject, or a motion of the subject.

[0212] In one or more scenarios, the behavior of the subject may comprise at least one of: scratching behavior, sleeping behavior, sleeping quality behavior, resting behavior, licking behavior, rubbing behavior, gnawing behavior, scooting behavior, defecation behavior, walking, running, rolling, digging, stair climbing, jumping, eating, drinking, urinating, defecating, and / or shaking behavior.

[0213] In one or more scenarios, the skin condition intervention effect may include at least one of: a grocery brand control food (CF) nutritional regime, or a therapeutically formulated test food (TF) nutritional regime.

[0214] In one or more scenarios, the at least one of: the grocery brand control food (CF) nutritional regime, or a therapeutically formulated test food (TF) nutritional regime, may be conducted over the period of time.

[0215] In one or more scenarios, the at least one ML model may comprise at least a full responder model. The full responder model may be configured to treat time as a categorical variable; and / or to assume a distinct treatment effect at each time point.

[0216] In one or more scenarios, the full responder model may comprise at least eighteen parameters.

[0217] In one or more scenarios, the at least one ML model may comprise at least a reduced responder model. The full responder model may be configured to treat time as a categorical variable; and / or to assume constant treatment effect at each time point.

[0218] In one or more scenarios, the reduced responder model may comprise at least ten parameters.

[0219] In one or more scenarios, the at least one ML model may comprise at least a support vector machine (SVM) classifier model.

[0220] In one or more scenarios, the SVM classifier model may be a one-class SVM classifier model.

[0221] In one or more scenarios, one or more methods may comprise receiving a first assessment of each of the subjects’ skin condition as a third data, the first assessment provided by one or more medical practitioners associated with each of the respective plurality of subjects over the period of time. The determining, via the at least one ML model, at least one of that the indicated skin condition intervention effect is relatively effective, or that the indicated skin condition intervention effect is relatively ineffective, for each of the plurality of subjects, may be based at least further in part, on the processing of the first data, the second data and the third data.

[0222] In one or more scenarios, the first assessment of each of the subjects’ skin condition may comprise, at least in part, one or more canine atopic dermatitis lesion index (CADLI) scores.

[0223] In one or more scenarios, each of the CADLI scores may be indicated by a real number within a range of zero to fifty.

[0224] In one or more scenarios, one or more methods may comprise receiving a second assessment of each of the subjects’ skin condition as a fourth data. The second assessment may be provided by one or more trained observers of the skin condition of each of the respective plurality of subjects over the period of time. The determining, via the at least one ML model, at least one of that the indicated skin condition intervention effect is relatively effective, or that the indicated skin condition intervention effect is relatively ineffective, for each of the plurality of subjects, may be based at least further in part, on the processing of the first data, the second data, the third data, and the fourth data.

[0225] In one or more scenarios, the second assessment of each of the subjects’ skin condition may comprise, at least in part, one or more Pruritus Visual Analog Scores (PVAS).

[0226] In one or more scenarios, each of the PVAS may be indicated by a real number within a range of zero to ten.

[0227] One or more computer-implemented methods / techniques / devices / systems / processes of determining a skin condition index for a plurality of subjects may be implemented. One or more methods may comprise receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. One or more methods may comprise identifying, as a second data, one or more specific behaviors exhibited by each of the plurality of subjects over the period of time based, at least in part, on the first data. One or more methods may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more methods may comprise processing, by the at least one ML model, the first data and the second data. One or more methods may comprise determining, via the at least one ML model, a skin condition for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. In one more scenarios, one or more of the aforementioned elements may be performed by one or more processors of one or more computing devices.

[0228] In one or more scenarios, the skin condition may be at least one of: an allergic condition, or a dermatological condition.

[0229] In one or more scenarios, the skin condition may be atopic dermatitis.

[0230] In one or more scenarios, the skin condition may be at least one of: an animal skin condition, or a human skin condition.

[0231] In one or more scenarios, the activity monitor may be at least one of: an animal collarmounted device, an accelerometer, and / or one or more human-body -mounted sensors.

[0232] In one or more scenarios, the at least some activity may be at least one of: a behavior of the subject, or a motion of the subject.

[0233] In one or more scenarios, the one or more specific behaviors may comprise at least one of: scratching behavior, sleeping behavior, sleeping quality behavior, resting behavior, and / or shaking behavior.

[0234] In one or more scenarios, the at least some activity may comprise data corresponding to at least one of: a maximum of an all-day sleep measurement, an intensity of one or more scratching bouts, a median of an all-day sleep measurement, or an intensity of one or more shaking bouts.

[0235] One or more computer-implemented methods / techniques / devices / systems / processes of evaluating a skin condition intervention effect for a plurality of subjects may be implemented. One or more methods may comprise receiving a first data from a plurality of activity monitors respectivelyassociated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. One or more methods may comprise, for each of the plurality of subjects, receiving an indicator as a second data identifying whether the dog is exhibiting signs of a skin condition, such as an allergic condition, food sensitivity, and / or a dermatological condition, either currently or over a period of time. One or more methods may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more methods may comprise processing, by the at least one ML model, the first data and the second data. One or more methods may comprise determining for a dog who is not part of either the first or second data, via the at least one ML model, at least one of: that the dog is exhibiting signs consistent with a skin condition, such as an allergic condition, food sensitivity or a dermatological condition, that the indicated skin condition therapy or intervention effect is relatively effective, or that the indicated skin condition therapy or intervention effect is relatively ineffective, for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. In one or more scenarios, one or more aforementioned elements may be performed by one or more processors of one or more computing devices.

[0236] In one or more scenarios, the skin condition intervention effect may include at least one of: a nutritional regime, a placebo regime, or a medicinal regime.

[0237] In one or more scenarios, the skin condition is at least one of: an allergic condition, or a dermatological condition.

[0238] In one or more scenarios, the skin condition may be atopic dermatitis.

[0239] In one or more scenarios, the skin condition may be at least one of: an animal skin condition, or a human skin condition.

[0240] In one or more scenarios, the activity monitor may be at least one of: an animal collarmounted device, an accelerometer, and / or one or more human-body-mounted sensors.

[0241] In one or more scenarios, the at least some activity may be at least one of: a behavior of the subject, or a motion of the subject.

[0242] In one or more scenarios, the behavior of the subject may comprise at least one of: scratching behavior, sleeping behavior, sleeping quality behavior, resting behavior, and / or shaking behavior.

[0243] In one or more scenarios, the skin condition intervention effect may include at least one of: a grocery brand control food (CF) nutritional regime, or a therapeutically formulated test food (TF) nutritional regime.

[0244] In one or more scenarios, the at least one of: the grocery brand control food (CF) nutritional regime, or a therapeutically formulated test food (TF) nutritional regime, may be conducted over the period of time.

[0245] In one or more scenarios, one or more methods may comprise receiving a first assessment of each of the subjects’ skin condition as a third data. The first assessment may be provided by one or more medical practitioners associated with each of the respective plurality of subjects over the period of time. The determining, via the at least one ML model, at least one of: that the indicated skin condition intervention effect is relatively effective, or that the indicated skin condition intervention effect is relatively ineffective, for each of the plurality of subjects, may be based at least further in part, on the processing of the first data, the second data and the third data.

[0246] In one or more scenarios, the first assessment of each of the subjects’ skin condition may comprise, at least in part, one or more canine atopic dermatitis lesion index (CADLI) scores.

[0247] In one or more scenarios, one or more methods may comprise receiving a second assessment of each of the subjects’ skin condition as a fourth data. The second assessment may be provided by one or more trained observers of the skin condition of each of the respective plurality of subjects over the period of time. The determining, via the at least one ML model, at least one of: that the indicated skin condition intervention effect is relatively effective, or that the indicated skin condition intervention effect is relatively ineffective, for each of the plurality of subjects, may be based at least further in part, on the processing of the first data, the second data, the third data, and the fourth data.

[0248] In one or more scenarios, the second assessment of each of the subjects’ skin condition may comprise, at least in part, one or more Pruritus Visual Analog Scores (PVAS).

[0249] One or more computer-implemented methods / techniques / devices / systems / processes of determining a skin condition index for a plurality of subjects may be implemented. One or more methods may comprise receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects. Each of the activity monitors may measure at least some activity conducted by the respective plurality of subjects over a period of time. One or more methods may comprise identifying, as a second data, one or more specific behaviors exhibited by each of the plurality of subjects over the period of time based, at least in part, on the first data. One or more methods may comprise inputting, into at least one machine-learning (ML) model, the first data and the second data. One or more methods may comprise processing, by the at least one ML model, the first data and the second data. One or more methods may comprise determining, via the at least oneML model, a skin condition for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data. In one or more scenarios, one or more of the aforementioned elements may be performed by one or more processors of one or more computing devices.

[0250] In one or more scenarios, the skin condition may be at least one of: an allergic condition, or a dermatological condition.

[0251] In one or more scenarios, the skin condition may be atopic dermatitis.

[0252] In one or more scenarios, the skin condition may be at least one of: an animal skin condition, or a human skin condition.

[0253] In one or more scenarios, the activity monitor may be at least one of: an animal collarmounted device, an accelerometer, and / or one or more human-body -mounted sensors.

[0254] In one or more scenarios, the at least some activity may be at least one of: a behavior of the subject, or a motion of the subject.

[0255] In one or more scenarios, the one or more specific behaviors may comprise at least one of: scratching behavior, sleeping behavior, sleeping quality behavior, resting behavior, and / or shaking behavior.

[0256] In one or more scenarios, the at least some activity may comprise data corresponding to at least one of: a maximum of an all-day sleep measurement, an intensity of one or more scratching bouts, a median of an all-day sleep measurement, and / or an intensity of one or more shaking bouts.

[0257] The subject matter of this disclosure, and components thereof, can be realized by instructions that upon execution cause one or more processing devices to carry out the processes and / or functions described herein. Such instructions can, for example, comprise interpreted instructions, such as script instructions, e. ., JavaScript or ECMAScript instructions, or executable code, and / or other instructions stored in a computer readable medium.

[0258] Implementations of the subject matter and / or the functional operations described in this specification and / or the accompanying figures can be provided in digital electronic circuitry, in computer software, firmware, and / or hardware, including the structures disclosed in this specification and their structural equivalents, and / or in combinations of one or more of them. The subject matter described in this specification can be implemented as one or more computer program products, e.g., one or more modules of computer program instructions encoded on a tangible program carrier for execution by, and / or to control the operation of, data processing apparatus.

[0259] A computer program (also known as a program, software, software application, script, orcode) can be written in any form of programming language, including compiled or interpreted languages, and / or declarative or procedural languages. It can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, and / or other unit suitable for use in a computing environment. A computer program may or might not correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs and / or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, and / or in multiple coordinated files (e.g., files that store one or more modules, sub programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers that may be located at one site or distributed across multiple sites and / or interconnected by a communication network.

[0260] The processes and / or logic flows described in this specification and / or in the accompanying figures may be performed by one or more programmable processors executing one or more computer programs to perform functions by operating on input data and / or generating output, thereby tying the process to a particular machine (e.g., a machine programmed to perform the processes described herein). The processes and / or logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) and / or an ASIC (application specific integrated circuit).

[0261] Computer readable media suitable for storing computer program instructions and / or data may include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices (e.g., EPROM, EEPROM, and / or flash memory devices); magnetic disks (e.g., internal hard disks or removable disks); magneto optical disks; and / or CD ROM and DVD ROM disks. The processor and / or the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0262] While this specification and the accompanying figures contain many specific implementation details, these should not be construed as limitations on the scope of any invention and / or of what may be claimed, but rather as descriptions of features that may be specific to described example implementations. Certain features that are described in this specification in the context of separate implementations can also be implemented in combination in perhaps one implementation. Various features that are described in the context of perhaps one implementation can also be implemented in multiple combinations separately or in any suitable sub-combination. Although features may be described above as acting in certain combinations and / or perhaps even (e.g., initially) claimed as such, one or more features from a claimed combination can in some cases be excised fromthe combination. The claimed combination may be directed to a sub-combination and / or variation of a sub-combination.

[0263] While operations may be depicted in the drawings in an order, this should not be understood as requiring that such operations be performed in the particular order shown and / or in sequential order, and / or that all illustrated operations be performed, to achieve useful outcomes. The described program components and / or systems can generally be integrated together in a single software product and / or packaged into multiple software products.

[0264] Examples of the subject matter described in this specification have been described. The actions recited in the claims can be performed in a different order and still achieve useful outcomes, unless expressly noted otherwise. For example, the processes depicted in the accompanying figures do not require the particular order shown, and / or sequential order, to achieve useful outcomes. Multitasking and parallel processing may be advantageous in one or more scenarios.

[0265] While the present disclosure has been illustrated and described in detail in the drawings and foregoing description, the same is to be considered as illustrative and not restrictive in character, it being understood that only certain examples have been shown and described, and that all changes and modifications that come within the spirit of the present disclosure are desired to be protected.

Claims

CLAIMSWhat Is Claimed Is :

1. A computer-implemented method of evaluating a intervention effect and / or a change in a condition for a plurality of subjects, the method comprising:(a) receiving a first data from a plurality of activity monitors respectively associated with the plurality of subjects, each of the activity monitors measuring at least some activity conducted by the respective plurality of subjects over a period of time;(b) for each of the plurality of subjects, receiving a second data identifying one or more indicators of at least one of: a life stage or a life-quality condition for which each of the plurality of subjects received the at least one of: the therapy, the nutrition, or the intervention effect over at least some portion of the period of time;(c) inputting, into at least one machine-learning (ML) model, the first data and the second data;(d) processing, by the at least one ML model, the first data and the second data; and(e) determining, via the at least one ML model, at least one of: the intervention effect is relatively effective, or that the intervention effect is relatively ineffective, for each of the plurality of subjects, based at least in part, on the processing of the first data and the second data, wherein at least steps (c)-(e) are performed by one or more processors of one or more computing devices.

2. The method of claim 1, wherein the life-quality condition is at least one of: anxiety, limited mobility, a gastro-intestinal condition, an allergic condition, a dermatologic condition, a food sensitivity, a liver condition, a weight condition, a heart condition, a joint condition, a thyroid condition, a urinary condition, a glucose management issue, or cancer.

3. The method of any of the previous claims, wherein the life stage is at least one of: an adolescent stage, a reproductive stage, a mid-life stage, or a geriatric stage.

4. The method of any of the previous claims, wherein the plurality of subjects comprise one or more of: a human, or an animal.

5. The method of any of the previous claims, wherein the intervention effect is at least one of: a therapy, a nutrition, a medical intervention effect , or a physical intervention effect .

6. The method of any of the previous claims, wherein the intervention effect is at least one of: a placebo regimen, a palliative regimen, a wellness regime, or a therapeutic regimen.

7. The method of claim 2, wherein a dermatologic condition comprises at least atopic dermatitis.

8. The method of any of the previous claims, wherein the activity monitor is at least one of: an animal-body-mounted device, an accelerometer, a human-body-mounted sensor, an animal-wearable sensor, or a human-wearable sensor.

9. The method of any of the previous claims, wherein the at least some activity comprises at least one of: a behavior of the subject, a motion of the subject, audio data from the subject, heart rate date from the subject, temperature data of the subject, or breathing rate or a movement of the subject.

10. The method any of the previous claims, wherein the at least some activity comprises at least one of: licking, walking, running, trotting, cantering, galloping, pacing, ambling, limping, abnormal gait, vomiting, urinating, defecating, passing a hairball, straining to urinate, straining to defecate, digging, eating, drinking, chewing, grooming, playing, dressing, bathing, driving, exercising, cooking, scratching, sleeping, sleeping, resting, or shaking.1 1 . The method of any of the previous claims, wherein the plurality of activity monitors includes at least one location sensor, wherein the at least one location sensor provides at least some location data corresponding to the at least some activity.

12. The method of claim 5, wherein the nutrition comprises at least one of: a nutrition regimen not specifically formulated for the at least one of: the life stage or the life-quality condition, a nutrition regimen formulated for the at least one of: the life stage or the life-quality condition, or a nutrition regimen therapeutically formulated for the at least one of: the life stage or the life-quality condition.

13. The method of any of the previous claims, wherein the at least one ML model comprises at least a full responder model, the full responder model configured to: treat time as a categorical variable; and assume a distinct treatment effect at each time point.

14. The method of claim 13, wherein the full responder model comprises at least eighteen parameters.

15. The method of any of the previous claims, wherein the at least one ML model comprises at least a reduced responder model, the full responder model configured to: treat time as a categorical variable; and assume constant treatment effect at each time point.

16. The method of claim 15, wherein the reduced responder model comprises at least ten parameters.

17. The method of claim any of the previous claims, wherein the at least one ML model comprises at least a support vector machine (SVM) classifier model.

18. The method of claim 17, wherein the SVM classifier model is a one-class SVM classifier model.

19. The method of any of the proceeding claims, the method further comprising: receiving a first assessment of the at least one of: each of the subjects’ life stage, or each of the subjects’ life-quality condition as a third data, the first assessment provided by one or more evaluators associated with each of the respective plurality of subjects over the at least some portion of the period of time, wherein the determining, via the at least one ML model, at least one of: the intervention effect is relatively effective, or that the intervention effect is relatively ineffective, for each of the plurality of subjects, is based at least further in part, on the processing of the first data, the second data and the third data.

20. The method of claim 19, wherein the one or more evaluators associated with the each of the respective plurality of subjects comprises: a medical practitioner, a therapist, a pet caretaker, a veterinarian, or a computer vision algorithm.

21. The method of claim 19, wherein the first assessment of the at least one of: each of the subjects’ life stage, or each of the subjects’ life-quality condition comprises, at least in part, an assessment provided via one or more assessment tools.

Images