Search

CN-121481355-B - Quality detection data analysis method for near-term liquid medicine preparation

CN121481355BCN 121481355 BCN121481355 BCN 121481355BCN-121481355-B

Abstract

The invention relates to the technical field of intersection of biological medicine and data processing, and discloses a quality detection data analysis method of a near-expiration date liquid medicine preparation. The method comprises the steps of collecting transmission light intensity signals through a multi-wavelength optical sensing array, compressing the transmission light intensity signals at the edge side to generate a 6-dimensional low-dimensional spectrum feature vector, uploading the 6-dimensional low-dimensional spectrum feature vector to the cloud, performing dynamic track matching based on a layered hidden Markov model constructed by a historical full-life-cycle spectrum database, calculating a degradation risk index and triggering early warning. The system comprises a sensing array, an edge computing unit and a cloud collaborative analysis platform. According to the invention, through the edge-cloud cooperative architecture, communication and calculation expenses are obviously reduced while the early warning precision is ensured, the early warning of degradation of not less than 48 hours in advance is realized, and the error of an intervention window is effectively avoided.

Inventors

  • CHEN YINGMEI
  • WU QINGHUA
  • Zhao Lingmei
  • LI MEILING

Assignees

  • 中国人民解放军西部战区总医院

Dates

Publication Date
20260512
Application Date
20260108

Claims (10)

  1. 1. A method of analyzing quality detection data of a near-expiration date liquid pharmaceutical formulation, comprising: The method comprises the steps of acquiring transmission light intensity signals of liquid medicines at a plurality of preset wavelengths in real time through a multi-wavelength optical sensing array deployed outside a medicine storage container to form an original multi-channel spectrum time sequence data stream, acquiring environmental parameter time sequence data in real time through a temperature and humidity sensor and an illumination intensity sensor deployed in a medicine storage environment, and keeping the sampling frequency consistent with the transmission light intensity signals; performing edge side light feature compression on the original multichannel spectrum time sequence data stream to generate a low-dimensional spectrum feature vector, calculating the mean value and variance of environment parameter time sequence data in a current time window, and splicing the mean value and variance serving as a 2-dimensional environment feature vector with the low-dimensional spectrum feature vector to form an 8-dimensional fusion feature vector; Uploading the 8-dimensional fusion feature vector to a cloud collaborative analysis platform; in the cloud collaborative analysis platform, a layered hidden Markov model is built based on a full life cycle spectrum database of historical batch medicines, and the layered hidden Markov model is integrated with environmental parameter influence factors to realize collaborative modeling of spectrum characteristics and environmental factors; Carrying out dynamic track matching and deviation measurement on the 8-dimensional fusion feature vector of the current batch of medicines by using the layered hidden Markov model, and calculating to obtain a medicine degradation risk index; When the drug deterioration risk index exceeds a preset threshold, triggering a near-term deterioration early warning instruction and generating intervention suggestion information.
  2. 2. The method for analyzing quality detection data of a near-term liquid pharmaceutical preparation according to claim 1, wherein the multi-wavelength optical sensor array comprises at least 5 light emitting units and a shared photoelectric receiving unit, the at least 5 light emitting units emit monochromatic light with center wavelengths of 280 nm, 320 nm, 410 nm, 550 nm and 650 nm respectively, each light emitting unit is sequentially polled and lightened, and the shared photoelectric receiving unit synchronously records transmitted light intensity under corresponding wavelengths, wherein the sampling frequency is not lower than 10 times per second.
  3. 3. The method of claim 1, wherein performing edge-side lightweight feature compression on the raw multichannel spectral temporal data stream to generate a low-dimensional spectral feature vector comprises: Carrying out sliding window average filtering on the original transmitted light intensity sequence of each wavelength channel, wherein the window length is 30 sampling points; calculating normalized absorbance values of each wavelength channel in a current time window, wherein the normalized absorbance values are equal to the ratio of the transmitted light intensity to the initial reference transmitted light intensity under negative natural logarithm; And selecting the adjacent time window change rate of the normalized absorbance value, namely the absolute value of the first-order difference value, and combining the normalized absorbance value corresponding to the maximum 3 wavelength channels and the first-order difference value thereof into a 6-dimensional feature vector serving as the low-dimensional spectrum feature vector.
  4. 4. The method according to claim 1, wherein the full life cycle spectrum database of the historical lot of medicines stores a multi-wavelength normalized absorbance sequence measured daily under the same environmental condition from the day of factory delivery to the day of failure of each historical lot of medicines, and key degradation inflection point time stamp and main component degradation path identification determined in the acceleration stability test of the corresponding lot.
  5. 5. The method for analyzing quality detection data of a near-term liquid medicine preparation according to claim 1, wherein the layered hidden markov model is a layered hidden markov model, wherein bottom state nodes represent initial oxidation or hydrolysis reaction stages of medicines at a microscopic molecular level, middle state nodes represent progressive shift stages of macroscopic physicochemical properties such as turbidity and chromaticity, top state nodes represent mutation degradation stages close to a failure period, and transition probability matrixes and observation probability distribution parameters of the layered hidden markov model are obtained through offline training by using a full life cycle spectrum database of historical batches of medicines through a maximum likelihood estimation method.
  6. 6. The method for analyzing quality detection data of a near-term liquid pharmaceutical formulation according to claim 5, wherein the step of performing dynamic trajectory matching and deviation measurement on the low-dimensional spectral feature vector of the current batch of pharmaceutical products by using the hierarchical hidden markov model, and calculating a pharmaceutical degradation risk index comprises: inputting a low-dimensional spectrum characteristic vector sequence of the current batch of medicines into the layered hidden Markov model, and calculating the most probable state transition path of the current batch of medicines; extracting a first entry time predicted value of a top-level state node in the state transition path; comparing the first entry time predicted value with the current remaining effective period, and judging that the degradation risk exists if the first entry time predicted value is earlier than the current remaining effective period minus 72 hours; The drug deterioration risk index is a negative function of the time difference between the first entry time predicted value and the current remaining effective period, specifically an exponential negative function, and the expression is: ; Wherein the method comprises the steps of For a time constant of 12 hours, In order to be a risk of deterioration index, For the current remaining period of validity, For the first time of entering the time prediction value, the risk index increases exponentially with decreasing time difference, and the smaller the time difference is, the higher the risk index is.
  7. 7. The method according to claim 6, wherein the predetermined threshold is set according to the type of drug, the time difference between the predetermined threshold and the predetermined threshold is 48 hours for a bio-product type liquid drug, and the time difference between the predetermined threshold and the predetermined threshold is 24 hours for a chemical synthesis type liquid drug.
  8. 8. The method of claim 1, wherein the intervention advice information includes one or more of advice to prioritize use of the batch of drugs, initiation of review procedures, adjustment of stored temperature and humidity parameters, or isolation of treatment to be scrapped, and the generation rules of the intervention advice information are jointly determined based on the drug class, current stored environmental parameters, and a quantified level of degradation risk index.
  9. 9. The method for analyzing quality detection data of a near-term liquid pharmaceutical preparation according to claim 1, characterized in that the method further comprises: locally caching the last 7 days of original multichannel spectrum time sequence data stream at the edge equipment; When the degradation risk index returned by the cloud collaborative analysis platform exceeds a preset threshold, the edge equipment automatically retrieves the cache data and executes high-resolution local re-analysis, and the method specifically comprises the steps of adopting sliding window filtering with a window length of 10 sampling points, calculating a second-order difference value of a normalized absorbance value, retaining original characteristics of all 5 wavelength channels, and verifying the reliability of an early warning result; and if the re-analysis result is consistent with the cloud pre-warning, locking the delivery authority of the batch of medicines, and sending a locking instruction to the quality management system.
  10. 10. The method for analyzing quality detection data of a near-term liquid pharmaceutical preparation according to claim 1, wherein the temperature and humidity sensor has a measurement range of-20 ℃ to 60 ℃ and a relative humidity of 10% to 95%, and measurement accuracy of + -0.5 ℃ and + -3%, respectively, and the illumination intensity sensor has a measurement range of 0 to 10000lux and a measurement accuracy of + -5%; the environmental parameter time sequence data and the transmitted light intensity signal are synchronously sampled, 1 group of environmental parameter data is generated every 3 seconds, and the environmental parameter data and the spectrum data of the corresponding time window are associated one by one; The mean value and variance of the environmental feature vector are calculated through environmental parameter data of 30 sampling points in a current time window, and the mean value and variance are spliced with the 6-dimensional spectral feature vector according to the sequence of the temperature mean value, the temperature variance, the humidity mean value, the humidity variance, the illumination mean value and the illumination variance, namely the original 6-dimensional spectral feature vector to form an 8-dimensional fusion feature vector.

Description

Quality detection data analysis method for near-term liquid medicine preparation Technical Field The invention belongs to the technical field of intersection of biological medicine and data processing, and particularly relates to a quality detection data analysis method of a near-expiration date liquid medicine preparation. Background With the continuous perfection of drug safety supervision systems and the remarkable improvement of public health consciousness, quality risk management and control of liquid drug preparations in the near-term has become a key issue in the links of drug circulation and use. The liquid medicine is easily affected by multiple environmental factors such as temperature, illumination, oxidation, microorganism pollution and the like in the storage and transportation processes, and the physicochemical properties of the liquid medicine can be irreversibly deteriorated along with the approach of the effective period, such as main component degradation, pH deviation, particle precipitation or clarity reduction. Traditional quality detection relies on laboratory offline analysis, has long period and delayed response, and is difficult to support dynamic risk early warning for near-term medicines. In recent years, on-line monitoring technology based on multi-wavelength spectrum sensing is gradually introduced into a medicine quality monitoring scene due to the advantages of non-invasiveness and real-time property, and the molecular structure change and impurity generation trend of medicines can be comprehensively represented by fusing ultraviolet-visible-near infrared and other multi-band spectrum data. Early quality degradation identification of near-term liquid drugs is highly dependent on refined fusion analysis of high-dimensional spectral features. The technical direction aims at extracting a recessive degradation index which is strongly related to the stability of the medicine through collaborative modeling of multi-wavelength signals, so that an early warning mechanism is triggered before the physical performance is obviously deviated. The method is characterized by constructing an intelligent analysis model capable of capturing weak spectrum drift and nonlinear degradation modes so as to realize advanced prediction of key quality attributes. In the prior art, multi-wavelength spectral data is typically collected by edge sensing devices deployed at pharmacy, cold chain car or storage nodes, forming a continuous real-time data stream. However, limited by the computational resource and power consumption constraints of the edge devices, existing schemes are difficult to perform high-complexity multi-wavelength feature fusion and deep learning reasoning tasks locally and efficiently. In order to reduce the calculation load, a part of methods adopt a simplified model or single-band analysis to cause the loss of key degradation information, and other schemes upload the original data to a cloud for processing, but network delay and bandwidth limitation cause serious delay of early warning response, and the gold stem expectation of drug degradation is often missed. Particularly in the basic medical institution or the mobile distribution scene, the contradiction among the calculation force, the precision and the time effect is particularly prominent, and a technical scheme capable of realizing early warning of the quality degradation of the medicine with high precision and low delay near-term at the limited edge end of the resource is needed. Disclosure of Invention The invention provides a quality detection data analysis method of a near-term liquid medicine preparation, and aims to solve the technical problem that a real-time data flow in quality detection is delayed in early warning of near-term degradation and misses a key intervention window due to the fact that multi-wavelength fusion analysis cannot be completed due to calculation force limitation of edge equipment. The invention provides a quality detection data analysis method of a near-term liquid medicine preparation, which comprises the following steps: The method comprises the steps of acquiring transmission light intensity signals of liquid medicines at a plurality of preset wavelengths in real time through a multi-wavelength optical sensing array deployed outside a medicine storage container to form an original multi-channel spectrum time sequence data stream, acquiring environmental parameter time sequence data in real time through a temperature and humidity sensor and an illumination intensity sensor deployed in a medicine storage environment, and keeping the sampling frequency consistent with the transmission light intensity signals; performing edge side light feature compression on the original multichannel spectrum time sequence data stream to generate a low-dimensional spectrum feature vector, calculating the mean value and variance of environment parameter time sequence data in a current time window, and splicing the me