Biomarkers for predicting chemotherapy-induced neuropathy

Abstract

A prediction device is configured to execute a prediction method for predicting a risk of chemotherapy-induced peripheral neuropathy (CIPN) in a subject. The prediction method includes receiving sensory data including low-frequency vibration perception data (LF-VPD) (211) and running at least one predictive model on the sensory data to determine at least one risk variable indicative of the risk of the subject developing chemotherapy-induced peripheral neuropathy (214). The LF-VPD indicates a measured perception of vibration at one or more predefined locations on one or more limbs of the subject and represents a vibration energy that causes the subject to transition between perception and non-perception of the vibration for each of one or more predefined frequencies below 64 Hz.

Description

Technical Field

[0001] The present disclosure relates to the monitoring of individuals undergoing chemotherapy, particularly to techniques for predicting chemotherapy-induced peripheral neuropathy (CIPN).

[0002] Background Art Chemotherapy (CTX) is a type of cancer treatment that uses one or more anti-cancer agents (chemotherapeutic agents) as part of a standardized chemotherapy regimen. Chemotherapy may be administered with the aim of cure or with the aim of prolonging life or reducing symptoms (palliative chemotherapy). It is also used in the treatment of diseases other than cancer, such as autoimmune and inflammatory diseases such as rheumatoid arthritis, psoriatic arthritis, psoriasis, polymyositis, Crohn's disease, vasculitis, systemic lupus erythematosus, multiple sclerosis, and AL amyloidosis. Chemotherapy is also used in the pretreatment regimen before bone marrow transplantation.

[0003] Chemotherapy-induced peripheral neuropathy (CIPN) is a common and serious side effect of chemotherapy. CIPN is a progressive, persistent, and often irreversible condition characterized by pain, numbness, tingling, coldness, etc. in the hands and feet, which affects a significant proportion of patients undergoing chemotherapy. CIPN can cause long-term disabilities and may interfere with daily life. When the hands are affected, patients may have difficulty handling small objects, dressing and undressing, taking care of themselves, and preparing meals. A decrease in foot sensitivity affects the sense of balance and the ability to perceive foot blisters, which may in turn increase the risk of developing foot ulcers.

[0004] Many chemotherapy agents, including platinum-based anti-cancer agents (oxaliplatin, cisplatin, etc.), vinca alkaloids, epothirone (ixabepyrone, etc.), taxanes, proteasome inhibitors (bortezomib, etc.), and immunomodulatory drugs (thalidomide, etc.), are known to be associated with an increased risk of neurological disorders. The prevalence is generally high. For example, oxaliplatin has been shown to cause CIPN in 95% of patients, and 20% exhibit long-term (persistent) tactile dysfunction even after completion of chemotherapy. If the condition persists for one year after the end of treatment, it is considered chronic.

[0005] The primary goal of chemotherapy is to save patients' lives by reversing or eliminating cancer cells, but the adverse effects of chemotherapy are also recognized by healthcare professionals. Therefore, in most clinics, patients are monitored for symptoms of peripheral neuropathy during chemotherapy. Based on this monitoring, clinicians may modify the treatment regimen, for example, by reducing the dose, switching to a different chemotherapy agent, or discontinuing chemotherapy.

[0006] In current clinical practice, the symptoms of peripheral neuropathy are often quantified through patient questionnaires that focus on perceived neurological symptoms and the perceived decline in daily living activities. Many different scales are available for quantifying symptoms, including the NTI-CTCAE, NCI-CTC, DEB-NTC, Oxaliplatin Scale, and EORTC QLQ-CIPN20.

[0007] Self-report questionnaires are quick and inexpensive. However, because self-reporting is inherently subjective, it is susceptible to psychological biases such as misunderstandings, overestimations, or underestimations due to factors like the patient's mood and willingness to acknowledge the existence of a problem. Furthermore, results have been shown to vary depending on the scale, and there is no consensus among healthcare professionals on which scale is most appropriate. In addition, symptoms may appear and disappear during treatment, and the correlation between symptoms experienced during treatment and long-term symptoms is low. Therefore, it is difficult for clinicians to accurately identify patients who are at high risk of developing long-term neurological damage. In some patients, chemotherapy may be unnecessarily adjusted, leading to inadequate cancer treatment, while in others, the need for chemotherapy adjustments may be overlooked, potentially resulting in permanent nerve damage.

[0008] In addition to using questionnaires, clinical tests may also be performed. The evaluation of various clinical tests is described in the paper “Quantitative Sensory Testing at Baseline and During Cycle 1 Oxaliplatin Infusion Detects Subclinical Peripheral Neuropathy and Predicts Clinically Overt Chronic Neuropathy in Gastrointestinal Malignancies” by Reddy et al., published in Clinical Colorectal Cancer, Vol.15, No.1, 37-46 (2016). Here, quantitative sensory testing (QST) is applied to identify early biomarkers for the development of CIPN in patients treated with oxaliplatin. QST is a panel of diagnostic tests used to assess somatosensory function and includes a wide range of sensations. Skin sensory thresholds were measured using a Von Frey monofilament. Pain perception was measured by applying a sharp tip to the skin. Vibrational perception (also known as vibratory touch) was measured using a 64 Hz graduated Rydell-Seiffer tuning fork. Thermal examinations assessed sensitivity to and tolerance to extreme heat and cold. Manual dexterity and fine motor skills were also quantified. The study in the aforementioned paper identified the cold and heat threshold and skin threshold as promising early biomarkers for the onset of CIPN. No clear relationship was found between pain and vibration perception and CIPN.

[0009] It should be noted that predicting the risk of CIPN in patients is different from detecting the presence of peripheral neuropathy in patients. Detecting peripheral neuropathy involves determining the patient's pre-existing condition, for example, based on self-reporting and / or medical testing. There is no basis for the assumption that any input data for a physician's diagnosis of pre-existing peripheral neuropathy would also be useful in identifying patients at risk of developing peripheral neuropathy due to chemotherapy at a future point in time. For example, the prior art described above clearly avoids using biomarkers based on pain and vibration perception for such predictions.

[0010] There is a continuing need for technologies that can identify patients at high risk of developing CIPN based on one or more biomarkers obtained from measurements taken in patients before or during chemotherapy. These technologies should preferably be time-efficient, so as not to require patients to undergo lengthy trials and multiple tests.

[0011] Summary of the Invention The objective is to overcome, at least partially, one or more limitations of the prior art.

[0012] A further objective is to provide technology for identifying patients at risk of developing CIPN.

[0013] Another objective is to provide such techniques based on measurement data obtained about patients before or during chemotherapy.

[0014] Another objective is to provide such technologies that are non-invasive and time-efficient.

[0015] One or more of these objectives, and any further objectives that may become apparent from the following description, are at least partially achieved by the devices, methods, and computer-readable media described in the independent claims, the embodiments thereof as defined by each dependent claim.

[0016] A first aspect of this disclosure is a predictive device. The predictive device comprises a circuit configured to predict the risk of chemotherapy-induced peripheral neuropathy (CIPN) in a subject. The circuit is configured to receive input data comprising perceptual data representing measured perceptions of vibrations at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration represents the vibrational energy that causes the subject to transition between perception and non-perception of the vibration, and at least one of the one or more predetermined frequencies is less than 64 Hz. The circuit is further configured to operate at least one predictive model on the perceptual data to determine at least one risk variable that indicates the risk of the subject developing CIPN due to chemotherapy, and to generate predictive data based on at least one risk variable.

[0017] The first aspect is based on the surprising finding that a subject's measured perception of vibration at one or more vibrational frequencies below 64 Hz contains information about the subject's likelihood of developing chemotherapy-induced peripheral neuropathy at a future point in time. In other words, by operating a appropriately configured predictive model on one or more biomarkers representing such measured perception, also known as vibrotactile sensation, the risk of chemotherapy-induced peripheral neuropathy, or CIPN, can be predicted. The predictive device of the first aspect provides a long-desired technology for identifying patients at risk of developing CIPN. Furthermore, the diagnostic capability of the predictive device can be based on the patient's vibrotactile sensation quantified by measurements before or during chemotherapy, without relying on subjective self-reporting by the patient. Such vibrotactile sensation can be measured non-invasively and time-efficiently using existing equipment.

[0018] A second aspect is a computer-based prediction method. The prediction method includes receiving input data which includes perceptual data representing measured perceptions of vibrations at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration represents the vibrational energy that causes the subject to transition between perception and non-perception of the vibration, and at least one of the one or more predetermined frequencies is less than 64 Hz. The prediction method further includes running a prediction model on the perceptual data to determine at least one risk variable that indicates the subject's risk of developing peripheral neuropathy due to chemotherapy, and generating prediction data based on the at least one risk variable.

[0019] A third aspect is a computer-readable medium containing computer instructions that, when executed by a processor, cause the processor to execute the prediction method or any embodiment of the second aspect.

[0020] A fourth aspect is a prediction method. The prediction method includes determining perceptual data representing measured perceptions of vibrations at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration represents the vibrational energy that causes the subject to transition between perception and non-perception of the vibration, and at least one of the one or more predetermined frequencies is less than 64 Hz. The prediction method further includes operating the prediction device of the first aspect on the perceptual data to generate prediction data indicating the risk of the subject developing peripheral neuropathy due to chemotherapy.

[0021] Further purposes, aspects, technical effects, features, and embodiments may be revealed in the following detailed description, the attached claims, and drawings.

[0022] The embodiments will be described in more detail below with reference to the attached drawings. [Brief explanation of the drawing]

[0023] [Figure 1A] FIG. 1A is a schematic side view, in partial cross section, showing an exemplary device for measuring vibratory tactile sensation. [Figure 1B] FIG. 1B is a diagram showing a system for measuring vibratory tactile threshold (VPT). [Figure 1C] FIG. 2 is a diagram showing an example of a vibrogram provided by the system of FIG. 1A. [Figure 1D] FIG. 3 is a flowchart of an exemplary method for determining VPT. [Figure 2A] FIG. 4 is a block diagram of an exemplary device for CIPN prediction. [Figure 2B] FIG. 5 is a flowchart of an exemplary method for performing CIPN prediction using low frequency vibration perception data (LF-VPD). [Figure 2C] FIG. 6 is a diagram illustrating the determination and use of LF-VPD for a chemotherapy session. [Figure 3A] FIG. 7 is a functional block diagram of prediction modules with different numbers of prediction models. [Figure 3B] FIG. 8 is a functional block diagram of prediction modules with different numbers of prediction models. [Figure 4A] FIG. 9 is a functional block diagram of prediction modules with different classification functions. [Figure 4B] FIG. 10 is a functional block diagram of prediction modules with different classification functions. [Figure 4C] FIG. 11 is a functional block diagram of prediction modules with different classification functions. [Figure 5] FIG. 12 is a flowchart of a procedure for evaluating the validity of input data for CIPN prediction. [Figure 6] FIG. 13 is a flowchart of a clinical use example of a risk class provided by CIPN prediction. [Figure 7] FIG. 14 is a diagram showing an example of the use of a combination of prediction modules for CIPN prediction based on temporally separated input data. [Figure 8A] FIG. 15 is a ROC plot determined for an exemplary prediction module. [Figure 8B]This is a determined ROC plot for an exemplary prediction module. [Figure 8C] This is a determined ROC plot for an exemplary prediction module. [Figure 9] This is a block diagram of a machine capable of performing any method, procedure, function, or step described herein.

[0024] Detailed description of exemplary embodiments The embodiments will be described in more detail below, with reference to the accompanying drawings illustrating some, though not all, embodiments. In fact, the subject matter of this disclosure can be embodied in many different forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided to satisfy the legal requirements to which this disclosure applies. Similar numbers refer to similar elements throughout.

[0025] Furthermore, where possible, it will be understood that any advantage, feature, function, device, and / or operational aspect of any of the embodiments and / or intended embodiments described herein may be included in any of the other embodiments and / or intended embodiments described herein, and / or vice versa. In addition, where possible, terms expressed in the singular form herein also include the plural form unless expressly otherwise stated, and / or conversely, terms expressed in the plural form may also include the singular form. As used herein, “at least one” means “one or more,” and these phrases are intended to be interchangeable. Thus, the term “one (a, an)” means “at least one” or “one or more,” even when the expressions “one or more” or “at least one” are also used herein. When used herein, unless the context explicitly requires otherwise by explicit wording or implicit suggestion, the words “comprise,” “comprises,” and variations such as “comprising,” are used in a comprehensive sense, i.e., to identify the presence of the described features, and do not preclude the presence or addition of further features in various embodiments of the invention. Similarly, the expressions “according to” and “based on” in combination with a specified set of parameters, etc., are also comprehensive and do not preclude the presence or addition of further parameters.

[0026] As used herein, the terms “multiple,” “several,” and “numerous” are intended to mean having two or more elements. The term “and / or” includes any and all combinations of one or more of the related elements listed.

[0027] As used herein, “prediction” refers to the process of determining, at the present time, the likelihood that a subject will be in a particular condition at a future time.

[0028] As used herein, “chemotherapy” (CTX) refers to any therapy involving the administration of chemotherapeutic agents to a subject for any therapeutic purpose. CTX is not limited to cancer treatment but can also be applied to the treatment of other diseases such as autoimmune diseases and inflammatory diseases, and to preparation for bone marrow transplantation. Chemotherapeutic agents are of any type of neurotoxic substance. For example, chemotherapeutic agents may be cytotoxic or anti-cancer agents that inhibit mitosis or induce DNA damage.

[0029] As used herein, “chemotherapy-induced peripheral neuropathy” (CIPN) refers to the development of peripheral neuropathy in subjects as a result of chemotherapy. In this context, “peripheral neuropathy” refers to a condition resulting from damage to nerves other than those in the brain and spinal cord (peripheral nerves). Symptoms of peripheral neuropathy may include pain, numbness, tingling, coldness, and decreased sensation. Peripheral neuropathy can affect any peripheral part of the body, but is usually first noticed in the hands and / or feet.

[0030] In this technical field, it is known to measure the vibrotactile sensation of subjects. The measured vibrotactile sensation can be used by physicians as input for medical analysis to detect, for example, progressive sensory impairment in subjects due to vibration exposure or diabetes. In a simple example, the measurement is performed by placing a tuning fork on the skin and vibrating it at its resonant frequency (typically around 100 Hz), and having the subject report whether or not they perceive the vibration. It is also known to measure vibrotactile sensation using specialized devices known as vibrometers, biotessiometers, parethesiometers, or neurotessiometers. These devices are configured to measure the vibrotactile threshold (VPT) at one or more vibration frequencies.

[0031] Figure 1 is a schematic diagram of such an exemplary device (hereinafter referred to as "vibrometer") 1. The vibrometer 1 is configured to generate measurement data representing vibrotactile sensation at one or more frequencies for a body part BP of the person being examined (hereinafter referred to as "patient" or "subject"). Measurement data can be generated for any body part, including the hands and feet. In the illustrated example, the body part BP is the hand, and vibrotactile sensation is measured for the fingers. The vibrometer shown in Figure 1A can be configured, for example, as described in International Publication No. 2006 / 046901.

[0032] The vibrometer 1 comprises a housing 2 (partially shown) that encloses the electromechanical measurement system and defines the opening. The housing 2 can also serve to mechanically support a body part BP to improve patient comfort and to ensure that measurement data is formed in a consistent and repeatable manner. The measurement system includes a measurement device 3 having a vibrating probe 4 positioned to protrude through the opening to expose the distal probe end 4A. The measurement device 3 includes a vibrator (not shown), which is an electrodynamic device that vibrates when current or voltage is supplied, and may also include a force sensor (not shown). The probe 4 is coupled to the vibrator, thereby enabling it to be operated to impart longitudinal vibration to the probe 4, as indicated by the bidirectional arrow. The force sensor can be directly or indirectly coupled to the probe 4 to sense the force applied longitudinally to the probe end 4A. The vibrometer 1 further includes a feedback device 5 to inform the patient that they have sensed the vibration of the probe 4. The feedback device 5 may include any combination of buttons, keyboards, keypads, touchscreens, microphones, gesture recognition systems, etc. Other types of feedback devices are also possible. The command device 6 enables the vibrometer 1 to output instructions or commands to the patient or the operator of system 1. The command device 6 may include any combination of indicator lights, displays, speakers, etc. The control unit 7 operates the vibrometer 1 and, connected by wire or wireless, supplies control signals to the measuring device 3, supplies command signals to the command device 6, and receives feedback signals from the feedback device 5.

[0033] The control unit 7 is configured to perform an examination procedure to form measurement data. During the examination procedure, the control unit 7 operates the measuring device 3 to apply vibrations of different energies to the probe 4 at predetermined frequencies. The patient engages their finger, for example, their fingertip as shown, with the probe end 4A and, upon sensing the vibrations of the probe 4, provides instructions via the feedback device 5. Such measurements can be repeated for one or more additional frequencies, resulting in measurement data for multiple frequencies. During the examination procedure, the control unit 7 can monitor force signals to ensure that the force applied by the fingertip to the probe end 4A is appropriate.

[0034] Figure 1B shows an embodiment of the vibrometer 1 in which the feedback device 5 is a push-button handle. In the illustrated example, the measurement data is transferred from the vibrometer 1 to an analysis device 10, which can be any type of computing device, for example, by wire or wireless.

[0035] Figure 1C is a so-called viprogram formed for the patient examination procedure. In Figure 1C, the vertical axis represents vibrational energy. The horizontal axis is divided into predetermined frequencies: 8 Hz, 16 Hz, 32 Hz, 64 Hz, 126 Hz, 250 Hz, and 500 Hz in this example. For each of the predetermined frequencies, Figure 1C shows a curve 12 representing the vibrational energy applied to the probe 4 as a function of time. In the illustrated example, the curve 12 forms a zigzag pattern, which means that the vibrational energy increases or decreases. Simultaneously, feedback from the patient is recorded. The feedback may indicate the end of perception while the vibrational energy is gradually decreasing, and / or the start of perception while the vibrational energy is gradually increasing. Based on this feedback, the vibrational tactile threshold (VPT) can be calculated. In Figure 1C, the VPT is represented by a horizontal line 14, and thus corresponds to the vibrational energy. The calculation of VPT is well known in the art, so further details are omitted. For example, see the ISO standards for vibration tactile measurement given in ISO 13091-1 and 13091-2.

[0036] Alternatively, VPT can be represented as a so-called z-score, also known as a standard score. The z-score indicates how many standard deviations above (positive) or below (negative) the mean of the data points. For example, the z-score can be calculated to normalize VPT for parameters such as age and sex by the mean, which represents the mean of VPT measured at a predetermined frequency for a selected patient population. The selected population can be chosen, for example, to have the same age, sex, and biometric data (e.g., height, finger temperature) as the patients. Patients in the selected population can be chosen based on having normal (undamaged) sensitivity to vibration.

[0037] Figure 1D is a flowchart of an exemplary method 200 for determining one or more VPTs for a patient. Method 200 can be performed by the vibrometer 1 in combination with the analysis device 10. Alternatively, method 200 may be performed by the vibrometer 1 itself.

[0038] In step 201, vibrations are induced at one or more predetermined frequencies with varying vibrational energy at one or more predetermined locations on one or more limbs of the patient, for example, using probe 4 in Figure 1A. In step 202, feedback is received from the patient indicating whether or not they perceive the vibrations during step 201. Steps 201-202 can be performed as a limit method or a levels method. In the limit method, the stimulus (vibrational energy) is gradually increased or decreased (see 12 in Figure 1C), and the subject provides feedback when they first perceive the stimulus during the increase or when they cease to perceive it during the decrease. In contrast, in the levels method, thresholds below or above a threshold are repeatedly applied, and the subject is asked to provide feedback indicating whether or not they perceive each stimulus. As can be understood from the above, steps 201-202 are performed separately for each predetermined frequency and each predetermined location. In step 203, based on the feedback, at least one VPT is calculated for each frequency and location, representing the vibrational energy at the point when the patient experiences a transition from perception to non-perception or from non-perception to perception, or both. For example, if step 201 involves forming multiple instances of measurement data for frequency and location, note that multiple VPTs can be calculated for a single frequency and location. Step 203 may include calculating one or more z-scores based on the VPTs.

[0039] In variations, the measurements in steps 201-202 can be performed simultaneously on multiple locations and / or body parts. For example, the vibrometer may be equipped with multiple vibrating probes or plates that engage with two or more locations on a body part (e.g., multiple fingers on one hand) and / or one or more locations on two or more body parts (e.g., the same fingers on different hands). It is also conceivable to control the vibrating probes / plates to vibrate simultaneously at two or more frequencies.

[0040] Figure 2A is a schematic diagram of an exemplary predictive device 20 configured to predict the risk of a patient experiencing CIPN at a future point in time, based on input data 21. The predictive device 20 can be configured to operate according to any of the methods, processes, routines, or steps presented herein. In the illustrated example, the predictive device 20 comprises an input interface 20A for receiving input data 21, a predictive module 300, and an output interface 20B for providing predictive data 22 indicating the predicted risk of CIPN. The predictive data 22 can be output from the predictive device 20 for storage in memory, display on a display unit, further processing, etc. The predictive device 20 may be any type of computer device, including, but is not limited to, a desktop computer, laptop, tablet, or smartphone. The input data 21 includes vibration perception data (VPD) representing one or more measurements of the patient's vibration touch. Depending on the implementation, the VPD may include the measurement data described above, or parameter data derived from the measurement data, such as one or more VPTs, one or more z-scores, etc. The prediction device 20 may be part of the analysis device 10 (Figure 1B) or the vibrometer 1 (Figure 1B), or it may be a separate device. Interfaces 20A and 20B may be internal or external interfaces, and it is understood that the input data 21 can be provided by electronic transfer, manual input, or any combination thereof.

[0041] Figure 2B is a flowchart of an exemplary method 210 that can be performed by the prediction device 20 to generate prediction data 22. Method 210 is the result of important experiments that led to the surprising discovery that vibration tactile measurements performed at vibration frequencies below approximately 64 Hz can indicate the future risk of CIPN. Hereafter, VPDs obtained at one or more frequencies below 64 Hz will be referred to as "low-frequency VPDs" or LF-VPDs. As shown in the viprogram in Figure 1C, measurements at 8 Hz, 16 Hz, and 32 Hz yield LF-VPDs. Currently, it is being considered to perform CIPN predictions based on LF-VPDs for frequencies below 60 Hz, below 50 Hz, below 40 Hz, below 35 Hz, below 30 Hz, below 25 Hz, below 20 Hz, below 15 Hz, or below 10 Hz. For practical reasons, vibration frequencies may be further restricted to above 0.1 Hz or above 0.5 Hz.

[0042] Method 210 can be performed before and / or during chemotherapy. In step 211, one or more VPDs with frequencies below 64 Hz, i.e., LP-VPDs, are received. In addition to LP-VPDs, VPDs with frequencies above 64 Hz may also be received in step 211. As indicated by the dashed lines, Method 210 may further include optional steps 212-213. In step 212, patient data (PD) representing the patient is received. In step 213, therapy data (CTD) representing the chemotherapy prescribed to the patient (ongoing, planned, or historical) is received. Patient data and therapy data may be in the form of a set of parameter values. Further examples of PD and CTD are shown below. In the example in Figure 2A, VPDs are included in input data 21, along with optional PD and / or CTD.

[0043] In the following description, we assume that VPD is represented in the input data 21 in the form of one or more “perceptual values” that represent the patient’s measured vibratory tactile sensation. For example, the perceptual values ​​may be VPT or z-scores. If, otherwise, VPD is represented in the form of the measurement data described above, the method may include steps similar to step 203 in Figure 1D, which calculate one or more perceptual values ​​from the measurement data.

[0044] Method 210 further includes step 214, which involves running a predictive model for VPD to determine at least one risk variable indicating the risk of a patient developing peripheral neuropathy due to chemotherapy. If available, step 214 may also involve running the predictive model for PD and / or CTD to determine risk variables. Step 215 generates predictive data (see 22 in Figure 2A) based on the risk variables. In some embodiments, the predictive data is generated to include the risk variables. In some embodiments, the predictive data is generated by processing the risk variables, as further illustrated below with reference to Figures 3-4.

[0045] The usefulness of prediction method 210 will be further explained with reference to Figure 2C, which schematically shows an example of chemotherapy (CTX) prescribed to a patient. CTX is generally administered at regular intervals, commonly referred to as cycles. A cycle may consist of administering one or more chemotherapy agents over a period of one or more days, followed by a period of several days or weeks without treatment ("drug-free period"). The drug-free period allows normal cells time to recover from side effects such as transient peripheral neuropathy. Doses may be administered consecutively over a certain number of days, or every other day for several days. Some chemotherapy agents are most effective when administered consecutively over a certain number of days. In this disclosure, a cycle with one or more doses is referred to as a "therapy session" or "session." Any two sessions are separated by a drug-free period. In the example in Figure 2C, CTX consists of a time-series TS of six therapy sessions. CTX is typically administered in the form of a so-called chemotherapy regimen ("CTX regimen"), which is prescribed by the attending clinician and specifies the chemotherapy agents, their dosages, the frequency and duration of treatment sessions, the method of administration of the chemotherapy agents (injection, tablets, etc.), and other considerations.

[0046] Figure 2C also shows different uses of prediction method 210 for VPDs measured before and during CTX, indicated by dashed arrows I-IV. In Option I, method 210 is performed on VPDs measured before the first TS, i.e., before the start of CTX, to predict the risk of CIPN at a long-term limit (LTL) given a given period ΔT after the completion of CTX, which may be set to 6 months, 12 months, etc. Such long-term CIPNs should be avoided as they are likely to have a long-lasting effect on the patient. Option I allows clinicians to adjust the CTX regimen to avoid or mitigate long-term CIPNs. In Option II, method 210 is performed on VPDs measured during CTX, in this example before the third TS, to predict the risk of long-term CIPNs. Option II allows clinicians to follow up on the risk of long-term CIPNs during CTX and adjust the CTX regimen accordingly. In Option III, method 210 is performed on VPDs measured before the first TS, to predict the risk of CIPNs during CTX ("short-term CIPNs"). Option III allows clinicians to adjust the CTX regimen to reduce patient discomfort during CTX and also reduce the risk of long-term CIPN, which is likely to correlate to some extent with short-term CIPN. In Option IV, Method 210 is performed on VPDs measured during CTX, in this example before the second TS, to predict the late risk during CTX, in this example, the risk of CIPN before the fourth TS. Option IV, like Option II, allows clinicians to follow up and adjust the CTX regimen. Although not shown in Figure 2C, it is also conceivable to perform Method 210 on combinations of VPDs measured at different time points. Generally, such combinations of VPDs would be measurable before each of at least two sessions and / or at two or more time points between sessions. In one example, the combination includes a VPD measured before the first TS and a VPD measured before the second TS. In another example, the combination includes VPDs measured at different time points during a single drug-free interval.In further examples, the combinations include VPDs measured at different points in time during drug-free periods.

[0047] The accuracy of the predictive data (22 in Figure 2A) can be improved by including therapeutic data CTD in the input data (21 in Figure 2A) and running the predictive model for combinations of VPD and CTD (step 214 in Figure 2B). The CTD can represent a CTX regimen and may include one or more parameters representing, for example, chemotherapy agents, dose, or schedule. In an example, dose may be given as the cumulative dose administered during CTX. In a further example, schedule may be given as the number of sessions, the duration of each session, the duration of drug-free periods, or a combination of these. The CTD may also include the patient's CTX treatment history, such as the CTX regimen of previous CTX, the time since the last CTX, and the time since the last treatment session in the ongoing CTX. The use of both VPD and CTD has the additional benefit that clinicians can assess the risk of CIPN for different CTX regimens in a patient by changing the CTD.

[0048] Furthermore, it has been found that the accuracy of the predictive data (22 in Figure 2A) can be improved by including patient data PD in the input data (21 in Figure 2A) and running the predictive model for combinations of VPD and PD (step 214 in Figure 2B). Examples of parameters that may be included in PD are age, sex, physical characteristics, current body temperature, health status, medication status, and medical history. Physical characteristics may be one or more of the following: height, weight, BMI, obesity, etc. Current body temperature is the patient's body temperature at the time the VPD was measured, for example, the temperature of the body part where the VPD was measured. Health status is the patient's condition at the time the prediction was made. Health status may indicate, for example, whether the patient has pre-existing neurological disorders and / or any comorbidities such as diabetes, Parkinson's disease, or multiple sclerosis (MS). Medication status may indicate any medications prescribed to the patient other than chemotherapy agents. Medication history may also indicate the dosage of such medications. Medical history (past illnesses) may include, but is not limited to, any related illnesses, diseases, or disorders that the patient has previously suffered from, including neurological disorders. Medical history may also include information such as the time since the last onset of an illness. Furthermore, PD may include subjective data from questionnaires, such as classification according to one of the existing scales described in the background technology section. In addition, PD may include data from other types of measurements of the patient, such as cold and heat thresholds, skin thresholds, and pain perception.

[0049] Furthermore, please understand that predictions can be further improved by running predictive models on combinations of VPD, CTD, and PD.

[0050] Furthermore, improved prediction can be achieved by running the predictive model on VPDs measured at multiple vibration frequencies below 64 Hz. VPDs at different frequencies can be measured at the same location on the same limb or different limbs, or at different locations. In some embodiments, the limb may be the hand or foot, and the location may be the metatarsal head of the foot or the fingers of the hand. If the patient has unilateral neuropathy, the VPD can be measured on the unaffected side of the patient. Currently, it is believed that predictive accuracy is improved when the input data consists of VPDs measured on at least one foot of the subject compared to VPDs measured on other parts of the human body.

[0051] Predictive improvements can also be achieved by running predictive models on VPDs measured at multiple locations and / or on one or more limbs.

[0052] Therefore, improvements can generally be achieved and made possible by running predictive models on multiple perceptual values ​​in VPDs, which differ by at least one of the following: vibration frequency, location, or limb.

[0053] The effectiveness of prediction method 210 is further demonstrated in Figures 8A–8C, which are so-called ROC plots. The ROC plots were generated by applying prediction method 210 to input data from clinical studies that assessed patients receiving CTX for signs of CIPN based on either self-report (NTI-CTCAE scale) or measured z-scores. Patients were divided into two groups with respect to different drug combinations, all of which included oxaliplatin. In Figure 8A, curve 801 is an ROC curve for predicting long-term CIPN (6 months after completion of CTX, self-reported) based on VPD (z-scores at 4Hz, 8Hz, and 16Hz measured at the first metatarsal head of the right foot), CTD (drug combination, number of sessions, and cumulative dose), and PD (age). In Figure 8B, curve 802 is an ROC curve for predicting long-term CIPN (measured z-score 6 months after completion of CTX) based on VPD (z-scores at 4Hz, 8Hz, 16Hz, and 32Hz measured at the first metatarsal head of the right foot) and CTD (drug combination, cumulative dose). In Figure 8C, curve 803 is an ROC curve for predicting short-term CIPN (self-reported at any time during CTX) based on VPD (z-scores at 4Hz, 8Hz, and 16Hz measured at the first metatarsal head of the right foot), CTD (drug combination), and PD (sex).

[0054] Figures 8A–8C illustrate, as an example, that prediction method 201 can predict long-term and short-term CIPN with good accuracy. The examples in Figures 8A–8C are included solely to demonstrate the effectiveness of prediction method 201, and there is no intention to limit the input data used in these examples. The input data used for specific CIPN predictions for a particular CTX regimen is a trade-off between the time required to collect the input data and the accuracy of the resulting CIPN predictions. The selection of input data can be determined by examination based on clinical or simulation data.

[0055] Examples in Figures 8A to 8C are all given for CTX administered with oxaliplatin. However, corresponding results are expected for other neurotoxic substances. According to scientific literature, the neurotoxicity of chemotherapeutic drugs is mainly caused by six different mechanisms: (i) activation of immune cells, (ii) damage to mitochondrial DNA transcription, (iii) changes in ion channel activity, (iv) damage to the myelin sheath, (v) disruption of microtubules, and (vi) metabolic changes in astrocytes. This finding is known, for example, from the paper "Mechanisms of Chemotherapy-Induced Peripheral Neuropathy" by Zajaczkowska et al., published in Int. J. Mol. Sci. 2019, 20, 1451. The mechanisms may act in parallel and may amplify each other. Mechanism (i) promotes inflammation of neurons, while mechanisms (iii) to (vi) mainly cause changes in the excitability of peripheral neurons. Mechanism (ii) involves both inflammatory and cytotoxic effects. Existing chemotherapeutic agents with neurotoxic effects can be broadly classified into six distinct groups: (a) platinum-containing drugs, (b) taxanes, (c) immunomodulators, (d) vinca alkaloids, (e) epothirone, and (f) protease inhibitors. The neurotoxicity of each of these groups stems from one or more of the mechanisms (i) to (vi). Oxaliplatin belongs to group (a) and exhibits mechanisms (i), (ii), and (iii). Currently, predictive method 210 is considered applicable to any CTX involving the administration of a chemotherapeutic agent exhibiting at least one of the mechanisms of oxaliplatin and / or a chemotherapeutic agent known to result in CIPN. Therefore, method 210 should be applicable to all other agents in group (a), including but not limited to carboplatin or cisplatin. Group (b) exhibits mechanisms (i), (ii), (iii), and (iv), and is thus closely similar to group (a) in terms of mechanism. Therefore, method 210 should be applicable to all drugs in group (b). Groups (c) to (f) all meet the criterion of sharing at least one mechanism with group (a) and / or being known to result in CIPN.Therefore, method 210 should be applicable to all drugs in groups (c) to (f).

[0056] Figure 3A is a block diagram of an exemplary prediction module 300 for use with the prediction device 20 of Figure 2A. In the illustrated example, module 300 is configured to receive and process input data in the form of VPD 301, and optionally PD 302 and CTD 303, in order to generate prediction data 351. The prediction data 351 includes a classification of CIPN risk, indicated by C. Classification C can include, for example, a representation of a risk class among several default risk classes, such as high (high risk of CIPN), low (low risk of CIPN), or a probability value for each default risk class. The illustrated prediction module 300 includes an optional preprocessing unit 30, which can be configured to adjust the input data before it is processed by the prediction model 32. For example, the preprocessing unit 30 can encode the input data into a format required by the model 32 and / or generate specific parameters from the input data. The preprocessing unit 30 can also remove outliers from the input data and / or perform validation checks on the input data, and optionally request further input data based on the validation checks (see Figure 5 below). The predictive model 32 is configured to generate one or more risk variables 331, shown in R. The risk variables 331 may indicate the risk of a patient developing low-term or short-term CIPN. Each risk variable can be given as a numerical value. Module 300 includes an evaluation unit 34 that processes the risk variables R to generate classification C. In one example, the evaluation unit 34 compares the risk variables to one or more thresholds. Examples of processing by the evaluation unit 34 are shown below with reference to Figures 4A to 4C. In some embodiments, the evaluation unit 34 may be omitted.

[0057] The prediction model 32 can define any functional relationship between the risk variable and a variable in the input data or a variable given by the input data. In some embodiments, the prediction model is linear to simplify the calculations. The linear model is

number

[0058] As is well known in the field, linear models use weight coefficients w i To determine the values, the model is calibrated or "trained" based on reference data. Once the weight coefficients are determined, the linear model is applied to module 300 as the predictive model 32. The determination of the weight coefficients can be performed by using any conventional mathematical method for predictive analysis of multivariate data, including, but is not limited to, partial least squares discriminant analysis (PLS-DA), sparse PLS-DA, decision trees (e.g., classification regression trees (CART), random forest classifiers, etc.), support vector machines (SVM), k nearest neighbors (k-NN), naive Bayes classification, etc.

[0059] Alternatively, the predictive model 32 may be a machine learning-based model (MLM), for example, an artificial neural network. Such a model also includes coefficients determined by training the model on reference data. Training can be performed by any conventional method.

[0060] It is understood that the reference data can be selected considering the discrimination or classification made possible by the risk variable. For example, to determine a set of predefined categories, the reference data should include patients classified into the set of categories. Categories may be mutually exclusive, but this is not required. Any number of categories can be used. In one example, the categories are "long-term CIPN" and "no CIPN". In another example, the categories are "short-term CIPN" and "no CIPN". In yet another example, the categories are different severities of CIPN, given by, for example, one of the existing scales described in the background technology section. In yet another example, the categories are "low risk of CIPN" and "not low risk of CIPN". In yet another example, the categories are "high risk of CIPN" and "not high risk of CIPN". In yet another example, the categories are different symptoms of CIPN, such as "numbness" and "pain".

[0061] Figure 3B is a block diagram of a modified version of the prediction module 30 in Figure 3A. Module 300 comprises a first prediction model 32 and a second prediction model 32'. The first model 32 is configured to generate one or more first risk variables 331, denoted by R, and the second model 32' is configured to generate one or more second risk variables 331', denoted by R'. The evaluation unit 34 is configured to determine the classification by an integrated evaluation of R and R'. Such an integrated evaluation can be performed in parallel, sequentially, or both. By performing an integrated evaluation using two models, the accuracy of the prediction can be improved. This improvement is not only due to the effect of providing additional risk variables, but also because the use of two models allows the models to be optimized for different types of discrimination, thereby improving both the sensitivity and specificity of module 300. More than two prediction models can also be used.

[0062] The first model 32 and the second model 32' may differ in the variables included in each model. The variables may, but do not need to, overlap at least partially between models 32 and 32'. Preferably, both models 32 and 32' operate on at least one variable representing LF-VPD. The number of variables (n) may differ between models 32 and 32'. Additionally or alternatively, models 32 and 32' may differ in the discrimination or classification enabled by the risk variable. Additionally or alternatively, models 32 and 32' may differ in the mathematical model used to calibrate or train each model. Additionally or alternatively, models 32 and 32' may be of different types, such as a linear model and a nonlinear model, or a linear model and an MLM.

[0063] As mentioned above, the predictive data may be classification C of the patient's risk. Classification C may represent one of a set of default risk classes. For example, the default risk classes can correspond to the categories mentioned above. For example, C may be either "long-term CIPN" or "no CIPN", either "short-term CIPN" or "no CIPN", a set of CIPN severity classifications, either "low risk of CIPN" or "not low risk of CIPN", either "high risk of CIPN" or "not high risk of CIPN", a set of symptom classifications, etc.

[0064] In some embodiments, the prediction module is configured to generate predictive data that indicates at least one of three default risk classes. This improves the usability of the predictive data for evaluating patients.

[0065] In some embodiments, at least three default risk classes include a first risk class associated with low risk, a second risk class associated with high risk, and a third risk class intermediate between the first and second risk classes. The first, second, and third risk classes may be associated with the risk of long-term CIPN, the risk of short-term CIPN, or both. The third risk class helps predictive data to distinguish between patients who need adjustments to their CTX regimen (second risk class) and those who simply need careful monitoring (third risk class). This can potentially lead to significant savings in healthcare resources.

[0066] In some embodiments, the prediction module 300 is configured to determine a risk category based on a comparison of a risk variable with at least one predetermined threshold and generate prediction data based on the risk category. In the example of FIGS. 3A-3B, such processing is performed by the evaluation unit 34. By a two-step procedure from the risk variable to the risk category and further from the risk category to the prediction data, it becomes possible to improve the accuracy of the prediction data. The advantages here are illustrated with reference to FIGS. 4A-4C.

[0067] In the example of FIG. 4A, the module 300 includes a prediction model 32 that generates a risk variable R indicating the risk that a patient will develop CIPN. The evaluation unit 34 is configured to define three determination steps 341-343 in which R is compared with first and second thresholds L1, L2 ("discrimination thresholds"). If it is found in step 341 that R ≤ L1, the patient is given the risk category "low", and the prediction data is set to the risk class 351A "low" (the first risk class). If it is found in step 342 that R > L2, the patient is given the risk category "high", and the prediction data is set to the risk class 351B "high" (the second risk class). Otherwise, if it is found in step 343 that L1 < R ≤ L2, the patient is given the risk category "doubt", and the prediction data is set to the risk class 351C "doubt" (the third risk class). It is understood that any of steps 341-343 may be implicit.

[0068] In the example in Figure 4B, module 300 includes two predictive models 32 and 32' that generate risk variables R and R', respectively. Risk variable R is generated to distinguish between risk categories "low" and "not low". Risk variable R' is generated to distinguish between risk categories "high" and "not high". Evaluation unit 34 is configured to define four decision steps 341-344 that compare R with a first threshold L1 and R' with a second threshold L2. Steps 341-344 correspond to logical combinations of the risk category determined for R (by comparison with L1) and the risk category determined for R' (by comparison with L2). If R > L1, the risk category is "low". If R ≤ L1, the risk category is "not low". If R' > L2, the risk category is "high". If R' ≤ L2, the risk category is "not high". In the example in Figure 4B, in step 341, if R results in the category "low" and R' results in the category "not high", the predicted data is set to risk class 351A "low". In step 342, if R results in the category "not low" and R' results in the category "not high", the predicted data is set to risk class 351C "doubtful". In step 343, if R results in the category "low" and R' results in the category "high", the predicted data is set to risk class 351C "doubtful". In step 344, if R results in the category "not low" and R' results in the category "high", the predicted data is set to risk class 351B "high".

[0069] The logical combination of risk categories determined for the two risk variables R and R' allows for higher sensitivity and specificity for the "low" and "high" risk classes in Figure 4B compared to Figure 4A.

[0070] It should be noted that the example in Figure 4B may also be performed by first determining the risk categories R and R', and then evaluating the risk categories for generating predictive data. In an alternative embodiment, the determination and evaluation of risk categories are performed alternately.

[0071] Furthermore, it should be noted that the logical combination of risk categories may differ from that shown in Figure 4B. For example, Model 32 may outperform Model 32' by setting the risk class "low" when R results in a "low" risk category, regardless of R's risk category. In another example, Model 32' may outperform Model 32 by setting the risk class "high" when R' results in a "high" risk category, regardless of R's risk category.

[0072] It should also be understood that further risk classes may be defined for the predicted data. Figure 4C shows an example different from Figure 4B, where, if R results in category "low" and R' results in category "high," the predicted data is set to risk class 351D "no class" in step 343. The use of a fourth risk class can further reduce the need for medical resources by reducing the number of patients assigned to risk class 351C "suspected." The rationale for risk class 351D is that if R and R' result in conflicting risk categories ("low" and "high"), something is likely wrong. Therefore, risk class 351D can be used to inform clinicians of potential errors, which will require them to review and, if necessary, update the input data.

[0073] The examples in Figures 4B-4C are also applicable to a single predictive model 32 configured to generate two or more risk variables. For example, such a predictive model can generate a first risk variable representing the risk of CIPN and a second risk variable which is the confidence score of the first risk variable. Similar to the processing of R,R' in Figures 4B-4C, the first risk variable can be compared to a first threshold to generate a risk category, and the second risk variable can be compared to a second threshold to generate a confidence category, and predictive data can be generated by a logical combination of risk categories and confidence categories.

[0074] Figure 5 is a flowchart of the validation procedure that can be performed by the preprocessing unit 30 (Figures 3A-3B). The procedure is performed each time the prediction module 300 receives input data. Step 310 analyzes the input data with respect to at least one requirement. Each requirement can be defined for VPD, PD, or CTD. In some embodiments, a violation of the VPD requirement occurs if 1) data is missing, for example in terms of the number or type of variables; 2) a deviation in the VPT or z score is detected; 3) force values ​​measured by the vibrometer and included in the input data are outside the specified range; 4) temperatures of body parts measured by the vibrometer and included in the input data are outside the specified range; 5) (assuming the input data indicates location / limb) the VPD is formed at the wrong location or limb; or 6) the confidence index formed by the vibrometer and included in the input data is below the limit. In some embodiments, a violation of the PD or CTD requirements may occur if one or more variables are judged to be poor or questionable, for example, if data is missing in terms of the number or type of variables, or if one or more variables are judged to be poor or questionable, for example, if the values ​​are outside a predetermined range, or if the combination of variable values ​​is unrealistic, such as if the ratio of a subject's height to weight is skewed. If the input data is determined to be valid in at least one requirement in step 310, the procedure proceeds to prediction by providing the input data to the prediction model 32. If the input data is determined to be in violation of a requirement or combination of requirements in step 310, the procedure proceeds to step 311, where counter p is incremented. In step 312, counter p is compared to the limit value pLim. If p is less than pLim, the method proceeds to step 313, where it outputs a request for new input data. The request may be output to a display and microphone, etc., and may also indicate the reason for the request in terms of the violated requirement, etc. Once new input data is received in response to the request, the procedure is repeated. If the procedure is repeated too many times and p ≥ pLim, the operator is notified in step 312 by setting the predicted data to risk class 351D "No class", for example, as shown in the figure.This causes the prediction operation to be terminated. It is understood that if the prediction was initiated in step 310, or if p ≥ pLim was determined in step 312, the counter p is reset. In a modified example, the limit value pLim may differ depending on the type of data. For example, a smaller limit value may be set for VPD (see method 200 in Figure 1D), which may need to be updated through subject measurements, compared to PD or CTD.

[0075] Note that, as is known in the art, if the prediction model 32 is configured to correct variables, for example, by using predicted values ​​instead, the prediction may proceed in step 310 even if certain variables are missing, poor, or questionable. However, one or more variables may be made mandatory in step 310, and the process may proceed to step 311 if these variables are missing, poor, or questionable. Furthermore, if the number of missing, poor, or questionable (non-mandatory) variables exceeds a limit, the process may proceed from step 310 to step 311.

[0076] Figure 6 shows an example of how the CIPN prediction described herein may be used in a clinical setting. The illustrated example assumes that the prediction data indicates one of the risk classes: "low," "high," or "suspected" (see 351A-351C in Figures 4A-4B). In step 600, basic or default range input data is obtained for the patient, and the prediction method 210 is performed on the default input data to generate prediction data. For example, a VPD can be obtained by performing measurements according to a default protocol. In addition to the VPD, the default input data may include a default set of variables for PD and / or CTD. If the prediction data generated in step 600 indicates a "low" risk class (step 601), the clinician may decide to subject the patient to routine monitoring for CIPN (step 602). Such routine monitoring may include taking and analyzing blood samples in connection with each treatment session and quantifying peripheral neuropathy symptoms using patient questionnaires. Otherwise, if the predictive data indicates a "high" risk class, the clinician may decide that the patient should undergo a clinical trial (step 604) to assess whether the patient is at high risk of developing CIPN (step 603). If a high risk is confirmed in the clinical trial, the clinician will re-evaluate CTX in step 605 and adjust the CTX regimen accordingly. As indicated by the dashed line, the clinical trial may be omitted. Otherwise, if the predictive data indicates a "suspected" risk class, the clinician may decide that predictive method 210 (step 606) should be performed on an expanded range of input data (step 603). The expanded input data may include additional variables of VPD, CTD, or PD and / or other variables compared to the default input data used in step 600. Step 606 may include obtaining the VPD by performing measurements according to the expanded protocol. In the example in Figure 6, step 606 assumes that predictive data indicating either a "low" or "high" risk class is generated.If the predictive data from step 606 indicates a "high" risk class, the clinician should re-evaluate the CTX (step 605). Otherwise, if the predictive data from step 606 indicates a "low" risk class, the clinician can place the patient in routine monitoring of CIPN (step 602).

[0077] Several advantages are easily understood from Figure 6. First, steps 600 and 606, respectively, allow for the early detection of patients who are susceptible to developing CIPN. Therefore, measures to mitigate or prevent CIPN can be taken earlier. This is particularly relevant to long-term CIPN, which can become chronic and cause unnecessary suffering. Second, by isolating patients in the intermediate risk class "suspected" using step 603, the time clinicians or physicians spend on potentially unnecessary re-evaluations of therapy can be reduced. Alternatively, the need for re-evaluation is quantified by step 606, allowing re-evaluation to be performed in a time-efficient manner by less medically qualified staff.

[0078] Figure 7 shows another use case. At time t=t1, input data 21 is acquired, and two prediction modules 300A and 300B operate on the input data 21. One prediction module 300A (MOD1) is configured to determine the risk of short-term CIPN, and the other prediction module 300B (MOD2) is configured to determine the risk of long-term CIPN. MOD1 and MOD2 operate on the input data 21, but it is understood that they can operate on at least partially different portions of the input data 21. MOD1 and MOD2 generate prediction data having one of three risk classes: "low," "high," and "suspected." MOD1 and MOD2 can be included in a single prediction device. For example, time t=t1 may be before the first session of CTX. Patients whose risk class is indicated as "low" or "high" by at least one of the modules can be managed as described in accordance with Figure 6. However, if a patient is identified as having a "suspected" risk class, the prediction device can advise the clinician to perform further CIPN predictions at a subsequent time point t=t2 during CTX, for example, before the next session. This advice is indicated by arrow 700 in Figure 7. In the illustrated example, CIPN prediction is performed at t=t2 by two other prediction modules 300C, 300D (MOD3, MOD4) operating on the input data 21 acquired at this point. MOD3 and MOD4, like MOD1 and MOD2, are configured to determine the risk of short-term and long-term CIPN, respectively, but generate prediction data with one of two risk classes: "low" and "high". This ensures that the separation of "suspected" into "low" or "high" (see step 607 in Figure 6) is temporally distributed across treatment sessions.

[0079] Figures 8A–8C are ROC plots of three different examples of predictive modules configured to represent one of two default risk classes, "high" and "low." Thus, the predictive module is a so-called binary classifier system. ROC plots are a well-known technique for demonstrating the diagnostic capability of a binary classifier system as its discriminant threshold is varied. ROC curves are created by plotting the true positive rate (TPR) against the false positive rate (FPR) at various threshold settings. TPR is also referred to as sensitivity. FPR is also called the miss rate and can be calculated as (1-specificity), where specificity is the true negative rate (TNR). ROC curves can be used to evaluate the performance of a binary classifier system and to set discriminant threshold values. In Figures 8A–8C, the diagonal dashed lines represent the results of random predictions, with the further the ROC curve is diagonally to the left, the better the prediction. In all of Figures 8A–8, each predictive module is shown to have good predictive capability. According to exemplary criteria, a method is considered practically useful if the sum of its sensitivity and specificity is at least 1.5. This criterion is met for most of the ROC curves 801-803 in Figures 8A-8C. The conditions and assumptions underlying each ROC plot have already been described above and will not be repeated here.

[0080] The structures and methods disclosed herein can be implemented by hardware or a combination of software and hardware. In some embodiments, the hardware comprises one or more software-controlled processors. Figure 9 schematically shows a prediction device 20 having an input interface 20A and an output interface 20B. Interfaces 20A, 20B can be implemented by hardware or a combination of software and hardware and can operate according to any standardized or proprietary protocol for wired or wireless data communication. The prediction device 20 further comprises a processing circuit 901 and computer memory 902. The processing circuit 901 may include one or more of the following: a CPU ("Central Processing Unit"), a DSP ("Digital Signal Processor"), a GPU ("Graphics Processing Unit"), a microprocessor, a microcontroller, an ASIC ("Application-Specific Integrated Circuit"), a combination of analog and / or digital discrete components, or any other programmable logic device such as an FPGA ("Field-Programmable Gate Array"). The control program 902A, which includes computer instructions, is stored in memory 902 and executed by processing circuit 901, which implements logic to perform any of the aforementioned methods, procedures, functions, operations, or steps. The control program 902A can be supplied to the prediction device 20 as a computer-readable medium 910, which may be a tangible (non-transient) product (e.g., a magnetic medium, optical disk, read-only memory, flash memory, etc.) or a propagating signal. As shown in Figure 9, memory 902 can also store control data 902B for use by processing circuit 901, such as the prediction model, the weight coefficients of the prediction model, the discrimination threshold, and the limit values.

[0081] While the present invention has been described in relation to the most practical and preferred embodiments at present, it should be understood that the present invention is not limited to the disclosed embodiments, but rather is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended claims.

[0082] Furthermore, while the diagrams show operations in a specific order, this should not be interpreted as requiring that these operations be performed in that specific order or sequentially, or that all illustrated operations be performed, in order to achieve the desired result. In some situations, parallel processing may be advantageous.

[0083] The following are clauses used to summarize some of the aspects and embodiments described above in the disclosure.

[0084] Clause 1. A predictive device comprising a circuit (901) configured to predict the risk of chemotherapy-induced peripheral neuropathy (CIPN) in a subject, wherein the circuit (901) receives input data comprising perceptual data (301) indicating measured perception of vibrations at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration represents the vibrational energy that causes the subject to transition between perception and non-perception of the vibration, and at least one of the one or more predetermined frequencies is less than 64 Hz; operates at least one predictive model (32,32') on the perceptual data (301) to determine at least one risk variable (331,331') indicating the risk of the subject developing peripheral neuropathy due to chemotherapy; and generates predictive data (351) based on the at least one risk variable (331,331').

[0085] Clause 2. The predictive device according to Clause 1, wherein the one or more default frequencies include at least two different frequencies less than 64 Hz.

[0086] Clause 3. A predictive device according to Clause 1 or Clause 2, wherein at least one of the one or more predetermined frequencies is 60 Hz or less, 50 Hz or less, 40 Hz or less, 35 Hz or less, 30 Hz or less, 25 Hz or less, 20 Hz or less, 15 Hz or less, or 10 Hz or less.

[0087] Clause 4. A predictive device according to any one of Clauses 1 to 3, wherein the perceptual data (301) includes a plurality of perceptual values ​​that differ by a predetermined frequency, a predetermined location, or at least one of the limbs of the subject.

[0088] Clause 5. A predictive device according to any one of Clauses 1 to 4, wherein the chemotherapy comprises a time series (TS) of sessions involving the administration of at least one chemotherapeutic agent, and the perceptual data (301) represents the measured perception of vibration by the subject prior to at least one session in the time series (TS).

[0089] Clause 6. The predictive device according to Clause 5, wherein the perceptual data (301) represents the measured perception of vibration by the subject at two or more points in time prior to each of at least two sessions in the session time series (TS) and / or at two or more points in time during one or more drug-free intervals between sessions in the session time series (TS).

[0090] Clause 7. A predictive device according to Clause 5 or Clause 6, wherein the perceptual data (301) represents the measured perception of vibration by the subject prior to at least the first session in the time series (TS) of the session.

[0091] Clause 8. The predictive device according to any one of Clauses 1 to 7, wherein the chemotherapy comprises the administration of at least one neurotoxic substance.

[0092] Clause 9. The predictive device according to any one of Clauses 1 to 8, wherein the chemotherapy comprises the administration of at least one chemotherapeutic agent from the group consisting of platinum-containing chemotherapeutic agents, taxanes, immunomodulators, vinca alkaloids, epothirones, and protease inhibitors.

[0093] Clause 10. A predictive device according to any one of Clauses 1 to 9, wherein the input data further includes a set of parameter values ​​(302, 303) representing the subject and / or the chemotherapy, and the circuit (901) is configured to determine the at least one risk variable by operating at least one predictive model (32, 32') for the perceptual data (301) and the set of parameter values ​​(302, 303).

[0094] Clause 11. The predictive device according to Clause 10, wherein the set of parameter values ​​(302,303) indicates one or more of the following: the age of the subject, the sex of the subject, one or more physical characteristics of the subject, the current body temperature of the subject, the health status of the subject, the medication status of the subject, the medical history of the subject, the history of chemotherapy treatment of the subject, the chemotherapeutic agents administered in the chemotherapy, the cumulative dose of the chemotherapeutic agents administered during the chemotherapy, the method of administration of the chemotherapeutic agents, and the schedule of the chemotherapy.

[0095] Clause 12. A predictive device according to any one of Clauses 1 to 11, wherein the one or more limbs include at least one of a foot or a hand.

[0096] Clause 13. The predictive device according to Clause 12, wherein the one or more designated locations consist of at least one of the metatarsal heads of the foot or the fingers of the hand.

[0097] Clause 14. The predictive device according to any one of Clauses 1 to 13, wherein the measured perception is represented in the perception data (301) by at least one vibration perception threshold (14), VPT, or at least one parameter value derived from at least one VPT (14).

[0098] Clause 15. The predictive device according to any one of Clauses 1 to 14, wherein the circuit (901) is further configured to generate a plurality of variables based on the input data, and the at least one predictive model (32) is configured to determine the at least one risk variable (331) by combining the plurality of variables using a plurality of predetermined weight coefficients.

[0099] Clause 16. The prediction device described in any one of Clauses 1 to 15, wherein the at least one prediction model (32,32') is one of a linear model or a machine learning-based model.

[0100] Clause 17. A predictive device according to any one of Clauses 1 to 16, wherein the at least one risk variable (331,331') includes a risk variable indicating the risk that the subject will develop chemotherapy-induced peripheral neuropathy lasting for at least six months after completion of chemotherapy.

[0101] Clause 18. A predictive device according to any one of Clauses 1 through 17, wherein at least one risk variable (331,331') indicates the risk of the subject developing chemotherapy-induced peripheral neuropathy during chemotherapy.

[0102] Clause 19. The predictive device described in any one of Clauses 1 to 18, wherein the predictive data (351) represents one of at least three default risk classes (351A, 351B, 351C).

[0103] Clause 20. The predictive device according to Clause 19, wherein the at least three default risk classes consist of a first risk class (351A) associated with low risk, a second risk class (351B) associated with high risk, and a third risk class (351C) intermediate between the first and second risk classes.

[0104] Clause 21. A predictive device according to any one of Clauses 1 to 20, wherein the circuit (901) is configured to determine a first category based on a comparison of at least one risk variable (331) with at least one predetermined threshold, and to generate predictive data (351) based on the first category.

[0105] Clause 22. A predictive device according to any one of Clauses 1 to 20, wherein the at least one risk variable includes a first risk variable and a second risk variable (331, 331'), and the circuit (901) is further configured to determine a first category based on the first risk variable (331), determine a second category based on the second risk variable (331'), and generate the predictive data (351) as a logical combination of the first category and the second category.

[0106] Clause 23. The predictive device according to Clause 22, wherein the circuit (901) is further configured to operate a first predictive model (32) on the input data (301) to determine the first risk variable, and to operate a second predictive model (32') on the input data (301) to determine the second risk variable.

[0107] Clause 24. A predictive device according to Clause 22 or Clause 23, wherein the first risk variable indicates a low risk of developing CIPN, and the second risk variable indicates a high risk of developing CIPN.

[0108] Clause 25. A predictive device according to any one of Clauses 1 to 24, wherein the circuit (901) is further configured to evaluate the input data with respect to a set of content requirements (310) to determine a validity score, and optionally output a request for further input data based on the validity score (313).

[0109] Clause 26. The predictive device according to Clause 25, wherein the circuit (901) is further configured to terminate predictive operation if the number of requests for the further input data exceeds a predetermined limit.

[0110] Clause 27. The prediction device according to any one of Clauses 1 to 26, wherein the prediction device further comprises an input interface (20A) and an output interface (20B) connected to the circuit (901), and the circuit (901) is configured to receive input data via the input interface (20A) and output prediction data (351) via the output interface (20B).

[0111] Clause 28. A computer-implemented prediction method comprising: receiving input data including perceptual data indicating measured perceptions of vibrations at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibrations represent the vibrational energy that causes the subject to transition between perception and non-perception of the vibrations, and at least one of the one or more predetermined frequencies is less than 64 Hz (211); operating a prediction model on the perceptual data to determine at least one risk variable indicating the subject's risk of developing peripheral neuropathy due to chemotherapy (214); and generating prediction data based on the at least one risk variable (215).

[0112] Clause 29. Computer-readable media containing computer instructions that cause the processing system to perform the methods described in Clause 28 when executed by the processing system.

[0113] Clause 30. A prediction method comprising: determining perceptual data indicating measured perception of vibrations at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration represents the vibrational energy that causes the subject to transition between perception and non-perception of the vibration, and at least one of the one or more predetermined frequencies is less than 64 Hz (200); and operating a prediction device according to any one of Clauses 1 to 27 on the perceptual data to generate prediction data indicating the risk of the subject developing peripheral neuropathy due to chemotherapy (210).

[0114] Clause 31. The method according to Clause 30, wherein determining the perceptual data (200) includes operating a vibration perception testing system to induce vibration energy varying at each of the one or more predetermined frequencies at each of one or more predetermined locations on one or more limbs of the subject (201), while receiving feedback (202) from the subject, and calculating the perceptual data based on the feedback (202).

Claims

1. A prediction device, The system includes a circuit (901) configured to predict the risk of chemotherapy-induced peripheral neuropathy (CIPN) in a subject, and the circuit (901) is configured to predict the risk of chemotherapy-induced peripheral neuropathy (CIPN) in a subject. The input data includes perceptual data (301) that represents the measured perception of vibration at one or more predetermined locations on one or more limbs of the subject, wherein for each of one or more predetermined frequencies, the input data includes perceptual data (301) that represents the vibration energy that causes the subject to transition between perceived and unperceived vibration. At least one predictive model (32, 32') is run on the aforementioned perceptual data (301) to determine at least one risk variable (331, 331') that indicates the risk of the subject developing peripheral neuropathy due to chemotherapy. Based on the aforementioned at least one risk variable (331, 331'), predictive data (351) showing the predicted risk of CIPN is generated. At least one of the one or more predetermined frequencies is less than 60 Hz, The at least one risk variable (331, 331') includes a risk variable indicating the risk that the subject will develop CIPN lasting for at least six months after the completion of the chemotherapy. Predictive device.

2. The one or more predetermined frequencies include at least two different frequencies less than 64 Hz. The prediction device according to claim 1.

3. At least one of the one or more predetermined frequencies is 50 Hz or less, 40 Hz or less, 35 Hz or less, 30 Hz or less, 25 Hz or less, 20 Hz or less, 15 Hz or less, or 10 Hz or less. The prediction device according to claim 1 or 2.

4. The perceptual data (301) includes a plurality of perceptual values ​​that differ by at least one of a predetermined frequency, a predetermined location, or the limb of the subject, A prediction device according to any one of claims 1 to 3.

5. The chemotherapy comprises a time series (TS) of sessions involving the administration of at least one chemotherapeutic agent, and the perceptual data (301) represents the measured perception of vibration by the subject prior to at least one session in the time series (TS). A prediction device according to any one of claims 1 to 4.

6. The perceptual data (301) represents the measured perception of vibration by the subject at two or more points in time prior to each of at least two sessions in the session time series (TS) and / or at two or more points in time during one or more drug-free intervals between sessions in the session time series (TS). The prediction device according to claim 5.

7. The perceptual data (301) represents the measured perception of vibration by the subject prior to at least the first session in the time series (TS) of the session. The prediction device according to claim 6.

8. The aforementioned chemotherapy includes the administration of at least one neurotoxic substance. A prediction device according to any one of claims 1 to 7.

9. The aforementioned chemotherapy includes the administration of at least one chemotherapeutic agent from the group consisting of platinum-containing chemotherapeutic agents, taxanes, immunomodulators, vinca alkaloids, epotilone, and protease inhibitors. A prediction device according to any one of claims 1 to 8.

10. The input data further includes a set of parameter values ​​(302, 303) representing the subject and / or the chemotherapy, The circuit (901) is configured to determine the at least one risk variable by operating the at least one prediction model (32, 32') on the perceptual data (301) and the set of parameter values ​​(302, 303). A prediction device according to any one of claims 1 to 9.

11. The set of parameter values ​​(302) represents one or more of the following: the subject's age, the subject's sex, one or more physical characteristics of the subject, the subject's current body temperature, the subject's health status, the subject's medication status, and the subject's medical history. The prediction device according to claim 10.

12. The set of parameter values ​​(303) represents one or more of the following: the subject's chemotherapy treatment history, the chemotherapy agents administered in the chemotherapy, the cumulative dose of the chemotherapy agents administered during the chemotherapy, the method of administering the chemotherapy agents, and the schedule of the chemotherapy. The prediction device according to claim 11.

13. The one or more limbs include at least one of a foot or a hand, A prediction device according to any one of claims 1 to 12.

14. The one or more predetermined locations consist of at least one of the metatarsal heads of the foot or the fingers of the hand. The prediction device according to claim 13.

15. The measured perception is represented in the perception data (301) by at least one vibration perception threshold (14), VPT, or at least one parameter value derived from the at least one VPT (14). A prediction device according to any one of claims 1 to 14.

16. The circuit (901) is further configured to generate a plurality of variables based on the input data, The at least one prediction model (32) is configured to determine the at least one risk variable (331) by combining the plurality of variables using a plurality of predetermined weight coefficients. A prediction device according to any one of claims 1 to 15.

17. The at least one predictive model (32, 32') is one of a linear model or a machine learning-based model. A prediction device according to any one of claims 1 to 16.

18. The aforementioned forecast data (351) represents one of at least three predetermined risk classes (351A, 351B, 351C). A prediction device according to any one of claims 1 to 17.

19. The at least three default risk classes consist of a first risk class (351A) associated with low risk, a second risk class (351B) associated with high risk, and a third risk class (351C) intermediate between the first and second risk classes. The prediction device according to claim 18.

20. The circuit (901) is configured to determine a first category based on a comparison between at least one risk variable (331) and at least one predetermined threshold, and to generate the prediction data (351) based on the first category. A prediction device according to any one of claims 1 to 19.

21. The at least one risk variable includes a first risk variable and a second risk variable (331, 331'), The circuit (901) is further configured to determine a first category based on the first risk variable (331), determine a second category based on the second risk variable (331'), and generate the prediction data (351) as a logical combination of the first category and the second category. A prediction device according to any one of claims 1 to 20.

22. The circuit (901) is further configured to operate a first prediction model (32) on the input data (301) to determine the first risk variable, and to operate a second prediction model (32') on the input data (301) to determine the second risk variable. The prediction device according to claim 21.

23. The first risk variable indicates a low risk of developing CIPN, and the second risk variable indicates a high risk of developing CIPN. The prediction device according to claim 21 or 22.

24. The circuit (901) is further configured to evaluate the input data with respect to a set of content requirements (310) to determine a validity score, and to selectively output a request for further input data based on the validity score (313). A prediction device according to any one of claims 1 to 23.

25. The circuit (901) is further configured to terminate predictive operation if the number of requests for the additional input data exceeds a predetermined limit. The prediction device according to claim 24.

26. The prediction device further comprises an input interface (20A) and an output interface (20B) connected to the circuit (901), and the circuit (901) is configured to receive the input data via the input interface (20A) and output the prediction data (351) via the output interface (20B). A prediction device according to any one of claims 1 to 25.

27. A prediction method implemented in a computer, Receiving input data which includes perceptual data representing the measured perception of vibration at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration energy that causes the subject to transition between perceived and unperceived vibration (211), (214) The predictive model is run on the aforementioned perceptual data to determine at least one risk variable that indicates the risk of the subject developing peripheral neuropathy due to chemotherapy, Based on the aforementioned at least one risk variable, predictive data showing the predicted risk of CIPN (215) Includes, At least one of the one or more predetermined frequencies is less than 60 Hz, The at least one risk variable (331, 331') includes a risk variable indicating the risk that the subject will develop CIPN lasting for at least six months after the completion of the chemotherapy. method.

28. When executed by a processing system, the computer instructions include causing the processing system to execute the method described in claim 27, Computer-readable media.

29. A prediction method, (200) Determining perceptual data that shows the measured perception of vibration at one or more predetermined locations on one or more limbs of a subject, wherein for each of one or more predetermined frequencies, the vibration energy that causes the subject to transition between perceived and unperceived vibration is determined, and at least one of the one or more predetermined frequencies is less than 64 Hz. (210) Operating the prediction device according to any one of claims 1 to 26 on the perceptual data to generate prediction data indicating the risk that the subject will develop peripheral neuropathy due to chemotherapy. including, Prediction method.

30. Determining the aforementioned perceptual data (200) The vibration perception testing system is operated (201) to induce vibration energy that changes at each of the one or more predetermined frequencies at one or more predetermined locations on one or more limbs of the subject, while receiving feedback (202) from the subject. (202) Calculating the perceptual data based on the aforementioned feedback and including, The prediction method according to claim 29.

Images