A gynecological patient discharge follow-up management system and method

Abstract

This invention discloses a follow-up management system and method for obstetric and gynecological patients after discharge. The system includes: a multidimensional specialty data acquisition and preprocessing module, a dynamic segmentation module for postpartum recovery stages, a terminal-side local differential privacy perturbation module, a maternal-infant dual-concern coupling assessment module, and a stage-adaptive follow-up strategy adjustment module. This invention uses cosine similarity template matching to segment postpartum recovery stages in real time, avoiding misjudging physiological decline as abnormal. At the patient terminal, a Laplace mechanism and a random response mechanism are used to perform ε-local differential privacy perturbation on hypersensitive physiological data, dynamically allocating a privacy budget to achieve source privacy compliance. A dual-concern coupling coefficient is constructed, and the degree of synergistic deviation between maternal recovery and newborn feeding is jointly assessed through area integral, dynamic time warping, and cross-terms. The follow-up frequency and questionnaire weights are dynamically adjusted based on the stage duration ratio and the coupling coefficient.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a follow-up management system for obstetric and gynecological patients after discharge. Background Technology

[0002] After delivery and discharge from the hospital, obstetric and gynecological patients enter the postpartum recovery period, which is characterized by a short cycle, drastic changes in physiological indicators, large individual differences, and the involvement of both maternal and infant health management. However, existing follow-up management systems are mainly designed for patients with chronic diseases, and their technical logic is based on the premise that the disease changes relatively slowly and the monitoring indicators are stable over a long period of time. Directly transplanting such systems to obstetrics and gynecology settings would result in a severe technological mismatch: First, postpartum physiological indicators of mothers decline rapidly due to physiological factors. This change is a normal recovery process rather than a pathological abnormal fluctuation. However, existing systems typically treat any deviation from the overall mean as abnormal, leading to misjudgments of numerous normal follow-up moments and unnecessary interventions. Second, obstetrics and gynecology follow-ups require simultaneous attention to two closely related health subjects: maternal recovery and newborn feeding. However, existing systems completely isolate the assessment of these two aspects, splitting the "maternal-infant dual health system" into two independent univariate systems and losing crucial coupling information. Furthermore, obstetrics and gynecology data is highly sensitive, involving highly sensitive information such as reproductive history, wound healing, and lactation function. Existing systems generally employ a centralized plaintext collection architecture, with patients' raw physiological data uploaded to traditional servers in an identifiable form, posing a risk of privacy leaks and failing to comply with the mandatory trend of the "Personal Information Protection Law of the People's Republic of China" and the regulatory trend of health and medical data security compliance. A search of existing technologies also revealed that although some systems focus on prenatal record keeping or introduce blockchain for evidence storage, none of them have solved the problem of dynamic segmentation of the rapid evolution of the postpartum period, achieved joint assessment of maternal and infant status, or implemented formal privacy protection mechanisms at the data collection layer. Summary of the Invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of this invention is to provide a follow-up management system and method for obstetric and gynecological patients after discharge, so as to solve the problems mentioned in the background art.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a follow-up management system for obstetric and gynecological patients after discharge. The system includes: a multi-dimensional specialty data acquisition and preprocessing module, a dynamic division module for the postpartum recovery stage, a terminal-side local differential privacy disturbance module, a maternal and infant dual-concern coupling assessment module, and a stage-adaptive follow-up strategy adjustment module.

[0005] The multidimensional specialty data acquisition and preprocessing module provided by this invention is used to acquire multidimensional specialty physiological data of obstetric and gynecological patients at each follow-up time after discharge, and to clean, interpolate and normalize the acquired data.

[0006] Specifically, the multidimensional specialty physiological data includes at least the following dimensions: (1) The lochia characteristics are graded into 1 to 5 ordered classification variables, where grade 1 is bloody lochia, grade 2 is serous lochia, grade 3 is white lochia, grade 4 is a small amount of serous fluid, and grade 5 is completely clean. (2) Wound healing score: The REEDA scale was used, which includes five sub-items: redness, edema, ecchymosis, secretions, and distance from the wound edge. Each sub-item is scored from 0 to 3 points, and the total score is scored from 0 to 15 points. Normalized healing degree W = 1 - (REEDA / 15). (3) 24-hour milk production, in milliliters, is automatically uploaded via Bluetooth by the smart breast pump or manually entered; normalized value M = actual milk production / 500, and truncated to the [0,1] interval; (4) Breast pain level: patients self-evaluate from 1 to 5, with 1 being no pain and 5 being severe pain; (5) Feeding frequency of newborns in 24 hours, in units of times; (6) The daily weight gain of the newborn, in grams per day, is automatically uploaded via Bluetooth baby scale; (7) The newborn's stool characteristics were quantified into 1 to 4 levels of ordered classification variables, where level 1 is meconium, level 2 is transitional stool, level 3 is breast milk soft stool, and level 4 is abnormal loose stool. (8) Edinburgh Postpartum Depression Scale score, total score 0-30; (9) Delivery method identifier: 0 represents natural childbirth, and 1 represents cesarean section.

[0007] The multidimensional specialist data acquisition and preprocessing module acquires data according to a preset acquisition frequency: once every 2 days in the first week postpartum, once every 3 days in the second to fourth weeks postpartum, and once a week in the fifth to sixth weeks postpartum.

[0008] For single instances of missing data, the module uses linear interpolation to fill the gaps. Let the time of the missing data be 1. The two most recent valid observation times before and after are respectively and Corresponding observation value and The interpolation formula is:

[0009] If data is missing for more than three consecutive times, the module marks that moment as an invalid follow-up and excludes it from subsequent stage discrimination and deviation calculation.

[0010] All continuous physiological indicators are subjected to min-max normalization before being input into subsequent modules, mapping the values ​​to the [0,1] interval.

[0011] The postpartum recovery stage dynamic segmentation module provided by this invention is used to determine the specific postpartum recovery stage in real time based on the temporal evolution trend of the patient's multidimensional specialized physiological data.

[0012] Specifically, the dynamic segmentation module for the postpartum recovery stage includes: a stage feature vector construction submodule, a postpartum stage template library storage submodule, a similarity matching and stage index generation submodule, a stage transition detection submodule, and a cold start processing submodule.

[0013] The stage feature vector construction submodule selects the lochia symptom grade L(t), the normalized lactation volume value M(t), and the normalized wound healing score value W(t) to construct the stage feature vector: The postpartum stage template library storage submodule is pre-loaded with a standardized recovery template library generated based on historical postpartum follow-up data. This template library is divided into a natural childbirth template library and a cesarean section template library according to the delivery method. Each template library contains four standard stage template vectors, corresponding to the four recognized sub-stages of postpartum recovery.

[0014] Taking the cesarean section template library as an example, the optimal value for its stage template vector is: Phase I (1-7 days postpartum, acute exudative phase): M1 = [3.2, 0.3, 0.4]; Phase II (8-14 days postpartum, milk production period): M2 = [2.1, 0.7, 0.7]; Phase III (15-28 days postpartum, functional recovery period): M3 = [1.4, 0.9, 0.9]; Phase IV (29-42 days postpartum, steady-state regression period): M4 = [1.1, 1.0, 0.95].

[0015] The similarity matching and stage index generation submodule calculates the current time feature vector V(t) and the template vectors for each stage. Cosine similarity: Where η = 1 × 10⁻ 6 This is a smoothing factor to prevent the denominator from being zero.

[0016] Take the stage number corresponding to the maximum similarity: Define the postpartum recovery stage index: Where PSI(t)∈(0,1], the larger the value, the later the recovery process is.

[0017] The stage transition detection submodule introduces a sliding window steady-state confirmation mechanism. Assuming a window length W=3, when all follow-up times within the window... When all values ​​are greater than or equal to the current stage threshold, and the moving average growth rate of PSI(t) is less than 1% / day, the patient is considered to have officially entered stage k.

[0018] The stage transition threshold is defined as follows: Thtrans ( k )=0.4* k The cold start processing submodule is used for initial stage assignment during the patient's first follow-up visit. The specific rules are as follows: ≤7 days postpartum calendar days default to Stage I, 8-14 days to Stage II, 15-28 days to Stage III, and ≥29 days to Stage IV. After the patient accumulates 3 valid follow-up visits, the system automatically switches to similarity matching mode.

[0019] The terminal-side local differential privacy perturbation module provided by this invention is deployed on the patient's mobile terminal to perform random perturbations on the raw physiological data to satisfy ε-local differential privacy, making it impossible to restore the individual's true values ​​in the data uploaded to the server. Specifically, it includes: a local differential privacy definition unit, a continuous data perturbation unit, a discrete-level data perturbation unit, a privacy budget dynamic allocation unit, and a bias-reduction estimation unit.

[0020] The local differential privacy definition unit stores the formal definition of ε-local differential privacy: for any two distinct input values ​​v and v' and any output y, the randomization algorithm M satisfies: Where ε is the privacy budget, the smaller the value, the stronger the privacy protection.

[0021] The continuous data perturbation unit employs a Laplace mechanism for continuous numerical data such as milk yield and newborn weight. Let the original true value be x, the sensitivity be Δf, and the privacy budget be ε, then the perturbation value x* is:

[0022] Where Lap(b) represents the Laplace distribution with scale parameter b. For data normalized to [0,1], Δf=1.0; for newborn weight (in g), Δf=100.

[0023] The discrete-level data perturbation unit employs a generalized random response mechanism for discrete ordered data such as lochia symptom levels, breast engorgement levels, and newborn stool symptom levels.

[0024] Let the size of the value space be |Z|, and the perturbation rule be: With probability Output the actual value; With probability Output any one of the other |Z|-1 values ​​with equal probability.

[0025] For a 5-point scale (|Z|=5), p = e^ε / (e^ε+4), q = 1 / (e^ε+4).

[0026] The privacy budget dynamic allocation unit dynamically allocates the privacy budget ε(t) based on the postpartum recovery stage index PSI(t) at the current moment: in =1.2, =0.6. This formula allows the privacy budget to decrease smoothly as the recovery process progresses, achieving an adaptive trade-off between accuracy and privacy.

[0027] The bias reduction estimation unit is deployed on the server side and is used to restore an approximately unbiased estimate of the original distribution from the group perturbation data.

[0028] For Laplace perturbations, the sample mean is an unbiased estimate of the original mean.

[0029] For a random response perturbation, let the frequency of observed class z after the perturbation be... Then the maximum likelihood estimator of the original class probability is: When the estimated value exceeds the range [0,1], it is truncated.

[0030] The maternal-infant dual-concern coupling assessment module provided by this invention is used to jointly assess the maternal recovery status and the newborn feeding status in each recovery stage, and to calculate the dual-concern coupling coefficient.

[0031] Specifically, it includes the following units: postpartum recovery deviation calculation unit, newborn feeding deviation calculation unit, dual attention coupling coefficient calculation unit, and missing dimension adaptive weighting unit.

[0032] The postpartum recovery deviation calculation unit uses wound healing score as the core indicator to calculate the patient's actual healing score sequence within stage j. Standard recovery curve compared to the same delivery method The area difference integral.

[0033] Let there be N_j valid follow-up visits within stage j, corresponding to times t1, t2, ..., t_{N_j}. Then, the postpartum recovery deviation is... for: in, ≥ 0, the larger the value, the more serious the deviation of the postpartum recovery process from the standard trajectory. The preset threshold of the unit is D_m_th=0.3, when A score >0.3 indicates delayed postpartum recovery.

[0034] The neonatal feeding deviation calculation unit uses the daily weight gain of newborns as the core indicator to calculate the actual daily weight sequence of newborns within phase j. Compared with the standard weight gain sequence of the same gestational age and sex The dynamic time warp distance.

[0035] Define DTW distance as: Where π is the optimal alignment path that satisfies the boundary conditions, monotonicity, and step size constraints.

[0036] Normalization yields the newborn feeding deviation: in As the maximum empirical value, this invention takes 25; the preset threshold value of the unit is D_n_th=0.35, when A value >0.35 indicates insufficient feeding of the newborn.

[0037] The dual-focus coupling coefficient calculation unit calculates the postpartum recovery deviation. Deviation from neonatal feeding Calculate the dual-concern coupling coefficient for stage j. : Wherein α, β, and γ are preset weighting coefficients, satisfying α + β + γ = 1. The preferred values ​​for the unit are: α = 0.35, β = 0.35, and γ = 0.30.

[0038] ∈[0,1], the larger the value, the higher the combined maternal and infant health risk. The unit presets a high-risk threshold of 0.6. When the value is greater than 0.6, the patient is automatically marked as high-risk and pushed to the medical staff for priority treatment.

[0039] The missing dimension adaptive weighting unit is used to handle scenarios where newborn data is missing. When the newborn weight sequence is incomplete, this unit executes the following logic:

[0040] Will The temporary value is the historical average deviation for this period; if there is no historical data, the default value of 0.2 is used. The weight γ of the interaction term is temporarily lowered to 0.1, while the weights α and β of the main effects are correspondingly increased to 0.45 and 0.45, respectively. Mark "Missing neonatal data, low confidence in coupling coefficient" in the follow-up report.

[0041] The stage adaptive follow-up strategy adjustment module provided by the present invention is used to dynamically adjust the follow-up frequency and the weight of follow-up content of patients according to the stage division result and the coupling evaluation result.

[0042] Specifically, it includes the following units: a follow-up frequency increase coefficient calculation unit, an adjusted follow-up number calculation unit, a follow-up content adaptive weighting unit, and a special follow-up task generation unit.

[0043] The follow-up frequency increase coefficient calculation unit defines the follow-up frequency increase coefficient \(k_j\) for the \(j\)th stage: Where: is the double-focus coupling coefficient for the \(j\)th stage; is the actual number of days in the \(j\)th stage, which is output in real time by the stage division module; is the reference stage duration, which is fixed at 7 days; \(\lambda\) is an adjustment factor, and its value range is 0.4 - 0.6. In the present invention, the preferred value is 0.5.

[0044] The adjusted follow-up number calculation unit calculates the adjusted follow-up number \(F_j'\) according to the preset basic follow-up number \(F_j\) for the \(j\)th stage and the follow-up frequency increase coefficient \(k_j\).

[0045] Among them, the basic follow-up number \(F_j\) is configured by the system default: Stage I: 2 times a week, ; Stages II and III: 1.5 times a week, ; Stage IV: 1 time a week, .

[0046] The formula for calculating the adjusted follow-up number is: Where \(\lfloor\ \rfloor\) is the floor function symbol. When \(F_j' > F_j\), the system automatically inserts a special follow-up in this stage; when \(F_j' = F_j\), the original follow-up plan is maintained; when \(F_j' < F_j\), the system only issues a prompt of "It is advisable to consider reducing the follow-up", and does not actively reduce the number of times.

[0047] The follow-up content adaptive weighting unit is used to dynamically adjust the weights of the special questions for parturients and the special questions for neonates in the follow-up questionnaire.

[0048] The system has built-in question banks for pregnant women and newborns, with initial base weights of [weight missing]. =0.5, =0.5.

[0049] when > When the weight adjustment formula is: when > When the weight adjustment formula is: When | - When |<0.05, the basic weight is maintained.

[0050] The special follow-up task generation unit determines the task type based on the special follow-up inserted when F_j'>F_j, according to the maximum deviation guidance principle: like > This special follow-up task is a special follow-up for postpartum women, which will be carried out by obstetric nurses, focusing on checking wound healing, lochia characteristics and psychological state; like > If the task of this special follow-up is a special follow-up for newborns, it will be carried out by pediatric nurses or breastfeeding instructors, with a focus on measuring weight, assessing feeding efficiency and stool characteristics; If | - If | ≤0.05, then this special follow-up task is a joint follow-up of mother and child, which will be carried out collaboratively by an interdisciplinary team.

[0051] The system provided by this invention adopts an end-to-cloud collaborative architecture.

[0052] The terminal-side components, deployed on the patient's mobile terminal, include: a data acquisition interface, a linear interpolation module, a normalization module, a local differential privacy perturbation module, a privacy budget dynamic allocation module, an encrypted transmission module, and a follow-up plan display module.

[0053] The cloud-based components deployed on the hospital follow-up server include: a perturbation data reception and decryption module, a bias estimation module, a postpartum stage template library, a stage division and transition detection module, a maternal-infant deviation and coupling coefficient calculation module, a follow-up strategy adjustment module, a special follow-up task distribution module, and a medical staff visualization dashboard module.

[0054] The data flow of the system is as follows: Raw physiological data is collected at the device, normalized, interpolated, and then input into the local differential privacy perturbation module. The perturbation module performs Laplace perturbation or random response perturbation based on the current privacy budget ε(t) to generate perturbation data. The perturbation data is encrypted and uploaded to the cloud. The cloud receives and decrypts the perturbation data and updates the patient's time-series database. The cloud calls the stage segmentation module to determine the current stage and PSI value through cosine similarity matching. The cloud calls the deviation calculation module to calculate the deviation values ​​respectively. , and The cloud side calls the follow-up strategy adjustment module to calculate k_j and F_j' and update the follow-up plan; the cloud side pushes the updated follow-up plan to the client side; the client side receives and displays the follow-up plan and prompts the patient to complete the next follow-up.

[0055] Throughout the entire process, the original, authentic data remains on the patient's terminal, while the cloud only stores the data after differential privacy perturbation.

[0056] Based on the above system, the present invention also provides an intelligent management method for follow-up visits of obstetric and gynecological patients after discharge, comprising the following steps: S1: Multidimensional Specialty Physiological Data Acquisition and Preprocessing The multidimensional specialist physiological data of patients at each follow-up time is obtained through the multidimensional specialist data acquisition and preprocessing module. The data includes at least the lochia symptom grade, wound healing score, 24-hour lactation volume, breast engorgement level, 24-hour feeding frequency of newborns, daily weight gain of newborns, stool symptom grade of newborns, Edinburgh Postnatal Depression Scale score and delivery mode identification. Data was collected according to the preset collection frequency: once every 2 days in the first week postpartum, once every 3 days in the second to fourth weeks postpartum, and once a week in the fifth to sixth weeks postpartum. For a single missing data point, linear interpolation is used to fill it in. For three or more consecutive missing data points, the time point is marked as an invalid follow-up and is not included in subsequent calculations. All continuous physiological indicators were subjected to min-max normalization and mapped to the [0,1] interval.

[0057] S2: Dynamic Division of Postpartum Recovery Stages and Generation of Stage Indices The following sub-steps are executed by dynamically dividing the postpartum recovery phase into modules: S201: Construct a stage feature vector, selecting lochia symptom grade L(t), normalized lactation value M(t), and normalized wound healing score W(t) to form V(t)=[L(t),M(t),W(t)]; S202: Call the pre-set postpartum stage template library. The template library is divided into natural childbirth template library and cesarean section template library according to the delivery method. Each template library contains 4 standard stage template vectors M1, M2, M3 and M4, which correspond to 1-7 days, 8-14 days, 15-28 days and 29-42 days postpartum, respectively. S203: Calculate the cosine similarity between V(t) and the template vector M_k at each stage; S204: Take the stage number k_max(t) corresponding to the maximum similarity and define the postpartum recovery stage index PSI(t) = k_max(t) / 4; S205: Stage transition detection. Assume a sliding window length W=3. When the k_max(t) of all follow-up times within the window is greater than or equal to the current stage threshold Th_trans(k)=0.4×k, and the moving average growth rate of PSI(t) is less than 1% / day, the patient is determined to have officially entered stage k. S206: Cold start processing. During the first follow-up, the initial stage is assigned according to the number of postpartum calendar days: ≤7 days postpartum is stage I, 8~14 days is stage II, 15~28 days is stage III, and ≥29 days is stage IV. After accumulating 3 valid follow-up data, the similarity matching mode is switched.

[0058] S3: Terminal-side local differential privacy perturbation The following sub-steps are performed on the patient's mobile terminal via the local differential privacy perturbation module on the terminal side: S301: For continuous numerical data such as milk production and newborn weight, a Laplace mechanism is used to generate perturbation values; S302: For discrete ordered data such as lochia symptom grades, breast tenderness grades, and newborn stool symptom grades, a generalized random response mechanism is adopted. S303: Dynamically allocate privacy budget based on the current postpartum recovery stage index; S304: Encrypt the disturbed data and upload it to the server; S305: After receiving the group disturbance data, the server performs bias-free estimation; S4: Maternal and Infant Dual Attention Coupling Assessment The following sub-steps are performed within each recovery phase using the maternal and infant dual-care coupling assessment module: S401: Calculate the postpartum recovery deviation. Using wound healing score as the core indicator, the area difference integral method was used to compare the actual healing score sequence within stage j. Standard recovery curve compared to the same delivery method ; S402: Calculate neonatal feeding deviation Using daily weight gain of newborns as the core indicator, the study compared actual weight sequences within stage j using dynamic time-normalized distance. Compared with the standard weight gain sequence of the same gestational age and sex ; S403: Calculate the dual-concern coupling coefficient ; S404: Perform adaptive weighting when newborn data is missing.

[0059] S5: Adjustment of Phase-Adaptive Follow-up Strategy The following sub-steps are executed through the phased adaptive follow-up strategy adjustment module: S501: Calculate the follow-up frequency increase coefficient for stage j; S502: Calculate the adjusted follow-up number based on the preset basic follow-up number F_j for stage j; S503: Dynamically adjusts the weight of content accessed in the questionnaire; the system has built-in question banks for maternal and neonatal patients. S504: Determine the type of special follow-up task based on the principle of maximum deviation; S505: The updated follow-up plan is pushed to the patient's app, and the patient receives and performs the next follow-up.

[0060] The beneficial effects of this invention are as follows: Targeting the rapid, phased characteristics of postpartum recovery in obstetric and gynecological patients, this invention constructs a dynamic phase division mechanism based on multidimensional feature template matching. By using cosine similarity to determine the specific recovery stage of the patient in real time, it avoids misjudging physiological declines as pathological abnormalities, ensuring a high degree of alignment between follow-up timing and the patient's actual recovery progress, significantly improving the phase adaptability of follow-up management. Simultaneously, this invention is the first to incorporate the mother's recovery status and the newborn's feeding status into the same assessment framework. It constructs a dual-concern coupling coefficient through area integral and dynamic time warping, and introduces cross-terms to quantify the special risks of asynchronous maternal and infant recovery, achieving a leap from "single-individual assessment" to "maternal-infant unit collaborative assessment."

[0061] This invention deploys a local differential privacy perturbation mechanism on the patient's terminal side. It employs Laplace's perturbation mechanism for continuous data and random response mechanism for discrete data, and dynamically allocates a privacy budget based on the recovery phase index. This ensures that the original, real data always resides on the terminal, while the cloud only stores the perturbated data, completely avoiding the risk of privacy leakage from the data collection source and fully complying with the requirements of the Personal Information Protection Law for sensitive health data. This mechanism does not rely on any hardware-level trusted environment, is cost-controllable, and is easy to deploy on a large scale. Furthermore, this invention constructs a follow-up frequency increase coefficient based on the phase duration ratio and the dual attention coupling coefficient, enabling flexible increases in the number of follow-up visits in non-integer multiples. It also dynamically adjusts the weight of follow-up questionnaire content based on the relative magnitude of maternal-infant deviation, precisely allocating follow-up resources to high-risk and asynchronous stages, effectively improving follow-up efficiency and the targeting of clinical interventions. Detailed Implementation

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1: Specific Implementation of Dynamic Division of Postpartum Recovery Stages This embodiment uses Ms. Zhang, a 29-year-old woman who underwent a cesarean section, as an example to explain in detail the calculation process of the stage division module.

[0064] Ms. Zhang underwent a cesarean section on January 15, 2026, and was discharged on January 20. The follow-up app collected data starting January 22 (7 days postpartum). Data from the 9th postpartum day (follow-up time t=9) is used for calculation.

[0065] Step 1: Feature Vector Extraction Patient uploaded data: lochia characteristics were "serous", corresponding grade L=2; 24-hour lactation volume was 260mL, normalized value M=260 / 500=0.52; REEDA wound healing score total score was 1 (edema 1 point, others 0 points), normalized healing degree W=1-(1 / 15)=0.93.

[0066] The eigenvector V(9) = [2.0, 0.52, 0.93].

[0067] Step 2: Template Similarity Calculation Cesarean section template library: M1=[3.2,0.3,0.4], M2=[2.1,0.7,0.7], M3=[1.4,0.9,0.9], M4=[1.1,1.0,0.95].

[0068] Calculate the cosine similarity (η=1e⁻) using M2 as an example. 6 ): Numerator: 2 × 2.1 + 0.52 × 0.7 + 0.93 × 0.7 = 4.2 + 0.364 + 0.651 = 5.215 Denominator: √(2²+0.52²+0.93²) × √(2.1²+0.7²+0.7²) = √(4+0.2704+0.8649) × √(4.41+0.49+0.49) = √5.1353 × √5.39 = 2.266 × 2.321 = 5.259 Sim2 = 5.215 / 5.259 = 0.992 Similarly, calculate Sim1=0.944, Sim3=0.958, and Sim4=0.902.

[0069] Step 3: Stage Identification and Stage Index The maximum similarity Sim2 was 0.992, corresponding to stage II. PSI = 2 / 4 = 0.5. The system determined that the patient was in the lactation stage, which was completely consistent with the obstetrician's clinical assessment.

[0070] If the existing global mean anomaly detection method is used, the milk production of 260 mL on the 9th postpartum day is significantly lower than the overall average for postpartum women in the hospital (approximately 450 mL), and would be misjudged as "highly abnormal," triggering unnecessary follow-up visits. This example fully demonstrates the superiority of the stage segmentation technology of the present invention.

[0071] Step 4: Stage transition detection The maximum similarity of the patient's two subsequent follow-up visits (day 11 and day 13) corresponded to stage II and stage III, respectively. The k_max(t) within the window was ≥2, and the moving average growth rate of PSI was 0.6% / day < 1% / day. The system officially determined that the patient had entered stage III on day 13 postpartum.

[0072] Example 2: Numerical Example of Local Differential Privacy Perturbation on the Terminal Side This embodiment takes lactation volume (continuous type) and lochia symptom grade (discrete type) as examples to fully demonstrate the terminal-side disturbance process.

[0073] Scenario 2.1: Laplace perturbation of milk production Patient Zhang's actual milk production on the 9th postpartum day was x = 260 mL, normalized x = 0.52. The current privacy budget ε(t) = 1.0 (according to the dynamic allocation formula, Stage II PSI = 0.5, ε = 1.2 - 0.6 × 0.5 = 0.9; the actual calculation should be 1.2 - 0.6 × 0.5 = 0.9, but for example, it is uniformly taken as 1.0). Sensitivity Δf = 1.0, scale parameter b = Δf / ε = 1.0.

[0074] Random noise is drawn from the Laplace distribution Lap(0,1), with an initial sampling value of η = 0.37. The perturbation value x* = 0.52 + 0.37 = 0.89, which is inversely normalized to 445 mL.

[0075] The server receives 445mL, but cannot distinguish whether this value represents the actual milk production or is a result of noise. Even if an attacker obtains all the data from the server, they cannot infer whether any patient's actual milk production is less than 300mL—this is the strict guarantee of ε-local differential privacy.

[0076] Scenario 2.2: Random Response Perturbation of Lochia Characteristics The patient's actual lochia grade is L=2 (serous). Current ε=1.0, value space |Z|=5.

[0077] p = e¹ / (e¹+4) = 2.718 / (2.718+4) = 2.718 / 6.718 = 0.4046 q = 1 / (e¹+4) = 1 / 6.718 = 0.1488 The app generates a random number rand=0.55. Since rand>p, the actual value is not output this time; instead, a level is randomly selected from {1,3,4,5}. Let the random selection result be level 4 (small amount of slurry).

[0078] The data uploaded to the server was "Lochia characteristics = 4". After collecting 1,000 cases of perturbation data, the hospital statisticians were able to restore the true proportion of the original grade 2 to approximately 32% using a debiasing estimator, but they had no idea of ​​Zhang's actual grade.

[0079] Scenario 2.3: Dynamic Allocation of Privacy Budget On the third day postpartum (Stage I, PSI≈0.25): ε=1.2-0.6×0.25=1.05, biased towards data availability.

[0080] On the 30th day postpartum (stage IV, PSI≈1.0): ε=1.2-0.6×1.0=0.6, indicating a bias towards privacy protection.

[0081] This dynamic allocation strategy ensures the effectiveness of clinical signals while gradually increasing the level of privacy protection as the recovery process progresses.

[0082] Example 3: Complete Calculation of the Coupling Coefficient of Maternal and Infant Dual Attention This embodiment uses Zhang's actual data in stage II (8-14 days postpartum) as an example to calculate the dual attention coupling coefficient.

[0083] Step 1: Maternal recovery deviation D_m,II The normalized values ​​of Zhang's wound healing score during the four follow-up visits in stage II were: t=8 days: 0.92, t=10 days: 0.94, t=12 days: 0.95, t=14 days: 0.96.

[0084] Standard recovery curve for cesarean section (stage II): S_std(t)=0.70+0.02×(t-8), i.e. [0.70,0.74,0.78,0.82].

[0085] The area difference integral is calculated using the trapezoidal method: Molecules (absolute area of ​​difference): (0.92-0.70)=0.22, (0.94-0.74)=0.20, (0.95-0.78)=0.17, (0.96-0.82)=0.14 Integral = (0.22 + 0.20) / 2 × 2 + (0.20 + 0.17) / 2 × 2 + (0.17 + 0.14) / 2 × 2 = 0.42 + 0.37 + 0.31 = 1.10 Denominator (Area under the standard curve): (0.70+0.74) / 2×2 = 1.44, (0.74+0.78) / 2×2 = 1.52, (0.78+0.82) / 2×2 = 1.60, total 4.56 D_m,II = 1.10 / 4.56 = 0.241 (Recovery is slightly better than the standard curve) Step 2: Neonatal feeding deviation D_n,II Daily weight gain (g / day) in newborns (female, 39 weeks gestation) during Stage II: [15, 22, 28, 31] WHO standard weight gain curve (39-week-old female infant, week 2): [18, 25, 30, 33] Calculate DTW distance (allowing for 1 day of scaling): Optimal alignment path: (15,18),(22,25),(28,30),(31,33) The square root distance is: √[(15-18)² + (22-25)² + (28-30)² + (31-33)²] = √(9+9+4+4) = √26 = 5.10 The empirical value of DTW_max is 25, so D_n,II = 5.10 / 25 = 0.204 Step 3: Double concern coupling coefficient Γ_II Take α=0.35, β=0.35, γ=0.30 Γ_II = 0.35×0.241 + 0.35×0.204 + 0.30×|0.241-0.204| = 0.08435 + 0.0714 + 0.30 × 0.037 = 0.15575 + 0.0111 = 0.16685 Γ_II = 0.167<0.6, which does not reach the high-risk threshold, indicating good maternal-infant synchronicity.

[0086] Example 4: Follow-up frequency adjustment and boundary condition handling Scenario 4.1: Adjustment of follow-up frequency under normal circumstances Following Example 3, the actual duration of Stage II is T_II = 7 days (8-14 days postpartum), T_base = 7, Γ_II = 0.167, and λ = 0.5.

[0087] k_II = 0.5 × 0.167 × (7 / 7) = 0.0835 Phase II baseline follow-up frequency F_II = ⌈1.5 × (7 / 7)⌉ = 2 times (rounded down to 1.5 times per week) The adjusted number of times F_II' = ⌊2 × (1+0.0835)⌋ = ⌊2.167⌋ = 2 times, maintaining the original plan.

[0088] Scenario 4.2: Extended phases lead to increased follow-up frequency If a patient's stage II is prolonged to 14 days due to complications, T_II=14, and Γ_II remains 0.167.

[0089] k_II = 0.5 × 0.167 × (14 / 7) = 0.167 F_II' = ⌊2 × (1+0.167)⌋ = ⌊2.334⌋ = 2 times (still 2 times, not exceeding the integer threshold); If Γ_II increases to 0.4, then k_II = 0.5 × 0.4 × 2 = 0.4, and F_II' = ⌊2 × 1.4⌋ = ⌊2.8⌋ = 2 times, which is still 2 times. The system is set to a maximum increase factor of ≤1.5, meaning F_II' can only increase a maximum of 3 times.

[0090] Scenario 4.3: Adaptive weighting under missing neonatal data If the patient has not uploaded the newborn's weight, the system will automatically perform adaptive weighting for the missing dimensions: D_n,j takes the historical average of 0.20 for this period (if there is no historical data, take the default value of 0.20); the cross term weight γ is temporarily lowered to 0.1, and α and β are raised to 0.45 and 0.45, respectively.

[0091] The recalculated Γ_II = 0.45×0.241 + 0.45×0.20 + 0.10×|0.241-0.20| =0.10845 + 0.09 + 0.0041 = 0.20255, which is slightly higher than the true value of 0.167. However, the follow-up report noted that "newborn data is missing, and the confidence level of the coupling coefficient is low".

[0092] Scenario 4.4: Adaptive Weighting of Follow-up Content If at a certain stage D_m=0.35 and D_n=0.20, then D_m>D_n, and the system executes: W_m_new = 0.5 × (1+0.35-0.20) = 0.5×1.15 = 0.575 W_n_new = 1 - 0.575 = 0.425 The proportion of questions specifically related to pregnant women in the questionnaire increased to 57.5%, while the proportion of questions specifically related to newborns was 42.5%.

[0093] Example 5: Optimal System Parameters for Large-Scale Deployment This invention employs a three-stage approach—literature extraction, simulation optimization, and clinical validation—to optimize core parameters, ultimately determining the default parameters as shown in the table below. Each medical institution can adjust these parameters according to local patient demographics through the system's backend access control interface, with complete auditable adjustment records.

[0094] Example 6: Software System Interface and Interaction Implementation This invention is delivered as an Android / iOS dual-platform APP with a Spring Boot microservice backend.

[0095] The patient-side app includes: The "Today's Measurement" module integrates a Bluetooth baby scale and a Bluetooth breast pump, automatically reading values ​​and performing differential privacy perturbation in the background; it also allows manual entry of lochia characteristics, stool characteristics, etc., with a pop-up message each time submission stating "Your data has been subjected to mathematical noise on your mobile device, and no institution can restore your true value."

[0096] The “Mother and Baby Diary” module displays historical measurement records in a calendar format (only the values ​​after perturbation are displayed, but the patient is informed that the original values ​​can be viewed locally on their mobile phone).

[0097] The “Follow-up Task” module displays the adjusted follow-up plan in card format, including the planned follow-up date, follow-up type (routine / specialized), and estimated time.

[0098] The healthcare web system includes: "Stage Monitoring Dashboard": Displays a pie chart showing the stage distribution of all managed mothers and the number of mothers in each stage in real time. Clicking allows you to drill down to view the individual stage index (PSI) change curve.

[0099] "Risk Warning List": Automatically filters patients with Γ>0.6 or D_m>0.3 or D_n>0.35, sorts them in descending order of coupling coefficient, and marks them as high risk in red.

[0100] "Follow-up plan configuration": Supports department directors to adjust weight parameters such as α, β, γ, and λ, and displays the changes in the number of follow-up visits after simulation parameter adjustment in real time.

[0101] "Privacy Compliance Audit": Records the privacy budget ε value and disturbance mechanism for each patient's upload, generating audit logs that comply with the requirements of the Personal Information Protection Law.

[0102] Example 7: Embedding a Legal Compliance Statement Before the system is first started and before each data collection, a pop-up window will display the "Health and Medical Data Authorization Collection Agreement" to clearly inform the patient: ① Your raw physiological data will have mathematical noise added on your mobile device, and the data uploaded to the server will be perturbed; ② No party (including hospitals, developers, and hackers) can use technical means to recover your true value from the disturbed data; ③ You have the right to withdraw your authorization at any time. After withdrawal, the system will stop collecting and delete the disturbance data stored in the cloud.

[0103] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A follow-up management system for discharged obstetric and gynecological patients, characterized in that, include: The multidimensional specialty data acquisition and preprocessing module is used to acquire multidimensional specialty physiological data of obstetric and gynecological patients at each follow-up time after discharge, and to clean, interpolate and normalize the acquired data. The postpartum recovery stage dynamic segmentation module is used to determine the specific postpartum recovery stage of the patient in real time based on the temporal evolution trend of the multidimensional specialized physiological data, and generate a postpartum recovery stage index. The terminal-side local differential privacy perturbation module is deployed on the patient's mobile terminal to perform random perturbations on the raw physiological data to satisfy ε-local differential privacy, so that the perturbation data uploaded to the server cannot restore the individual's true value. The maternal and infant dual-concern coupling assessment module is used to jointly assess the mother's recovery status and the newborn's feeding status in each recovery stage and calculate the dual-concern coupling coefficient. The phase-adaptive follow-up strategy adjustment module is used to dynamically adjust the patient's follow-up frequency and follow-up content weights based on the recovery phase, the dual attention coupling coefficient, and the actual duration of the phase.

2. The obstetrics and gynecology patient discharge follow-up management system as described in claim 1, characterized in that, The multidimensional specialist physiological data include at least: lochia characteristics grade, wound healing score, 24-hour milk production, breast engorgement grade, 24-hour feeding frequency of newborns, daily weight gain of newborns, stool characteristics grade of newborns, Edinburgh Postnatal Depression Scale score, and delivery method identification. The multidimensional specialist data acquisition and preprocessing module acquires data according to a preset acquisition frequency: once every 2 days in the first week postpartum, once every 3 days in the second to fourth weeks postpartum, and once a week in the fifth to sixth weeks postpartum. For single instances of missing data, linear interpolation is used to fill the gaps. Let the time of the missing data be t. The two most recent valid observation times are respectively and Corresponding observation value and The interpolation formula is: If data is missing for more than three consecutive times, that time point will be marked as an invalid follow-up and will not be included in subsequent stage judgments and deviation calculations.

3. The obstetrics and gynecology patient discharge follow-up management system as described in claim 1, characterized in that, The dynamic segmentation module for the postpartum recovery stage includes: a stage feature vector construction submodule, a postpartum stage template library storage submodule, a similarity matching and stage index generation submodule, a stage transition detection submodule, and a cold start processing submodule.

4. The obstetrics and gynecology patient discharge follow-up management system as described in claim 1, characterized in that, The terminal-side local differential privacy perturbation module deployed on the patient's mobile terminal includes: a local differential privacy definition unit, a continuous data perturbation unit, a discrete-level data perturbation unit, a privacy budget dynamic allocation unit, and a bias-free estimation unit.

5. The obstetrics and gynecology patient discharge follow-up management system as described in claim 1, characterized in that, The maternal and infant dual-concern coupling assessment module includes the following units: The calculation units include: postpartum recovery deviation calculation unit, newborn feeding deviation calculation unit, dual attention coupling coefficient calculation unit, and missing dimension adaptive weighting unit.

6. The obstetrics and gynecology patient discharge follow-up management system as described in claim 1, characterized in that, The phased adaptive follow-up strategy adjustment module includes: a follow-up frequency increase coefficient calculation unit, an adjustment follow-up number calculation unit, an adaptive weighting unit for follow-up content, and a special follow-up task generation unit.

7. A method for managing the follow-up after discharge of obstetric and gynecological patients, applied to the system described in any one of claims 1 to 6, characterized in that, Includes the following steps: S1: The multidimensional specialty physiological data of patients at each follow-up time is acquired through the multidimensional specialty data acquisition and preprocessing module, and then cleaned, interpolated and normalized. S2: The postpartum recovery stage dynamic segmentation module determines the patient's current specific postpartum recovery stage in real time based on the temporal evolution trend of the multidimensional specialized physiological data, and generates a postpartum recovery stage index. S3: The patient's mobile terminal performs random perturbation on the raw physiological data to satisfy ε-local differential privacy through the terminal-side local differential privacy perturbation module, generates perturbation data, and uploads it to the server; S4: The maternal and infant dual attention coupling assessment module is used to jointly assess the maternal recovery deviation and the newborn feeding deviation in each recovery stage, and calculate the dual attention coupling coefficient. S5: The phase-adaptive follow-up strategy adjustment module dynamically adjusts the patient's follow-up frequency and follow-up content weight based on the recovery phase, the dual attention coupling coefficient, and the actual duration of the phase, and pushes the updated follow-up plan to the patient.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of claim 7.

Images