1. Prediction of quality of medicines from manufacturing process timeseries data and spectra of incoming raw materials
- Author
-
Žagar, Janja and Mihelič, Jurij
- Subjects
spektri ,produkt ,laboratory analiysis ,spectra ,laboratorijska analiza ,kakovost ,prediction models ,podatki ,napovedni modeli ,serija ,batch ,data ,quality ,process ,product ,proces - Abstract
Eden pomembnejših ciljev farmacevtske industrije je pravočasna dostava zdravila z ustrezno kakovostjo do pacienta. Proizvodnja zdravil je visoko reguliran proces, kjer se za vsako serijo preverja kakovost vhodnih materialov, polprodukta in končnega produkta. Samo z vsemi analizami kakovosti znotraj definiranih mej je končna kakovost zdravila potrjena in se serija lahko sprosti na trg. Takšen proces zagotavljanja kakovosti je izredno zamuden zaradi analiz, ki se izvajajo predvsem laboratorijsko. Za izbrano zdravilo smo v sklopu doktorske disertacije raziskali, če lahko z uporabo zgodovinskih podatkov, ki jih med preverjanjem kakovosti in proizvodnjo zdravila zberemo, napovemo kakovost novih serij. Za ta namen smo zbrali podatke 1005 industrijsko proizvedenih serij in za vsako od teh serij 53 laboratorijskih analiz surovin, polprodukta in končnega produkta. Poleg tega pa še identifikacijske spektre surovin in časovne vrste ključnih procesnih parametrov. Podatki so bili zbrani iz več podatkovnih baz, ustrezno očiščeni in strukturirani, tako da smo jih lahko uporabili za napovedne modele. Z uporabo identifikacijskih NIR spektrov smo razvili napovedne modele za določanje kakovosti vhodnih materialov, s časovnimi vrstami procesa pa napovedne modele za optimizacijo ključnih proizvodnih korakov. Z vključitvijo vseh laboratorijskih analiz nam je uspelo razviti tudi napovedne modele za določanje kakovosti končnega produkta. Pokazali smo, da je možno zgolj z uporabo obstoječih zgodovinskih podatkov, ki jih industrija hrani za vsako serijo, razviti napovedne modele, ki bi nadomestili trenutne laboratorijske analize. S tem bi prihranili veliko časa, sprostili laboratorijske kapacitete in dostavili zdravilo na trg v znatno krajšem času. One of the most important aims within the pharmaceutical industry is the on-time delivery of medicine with expected quality to patients. Medicine production is a highly regulated process that requires quality of incoming materials, intermediates, and the final product to be verified for every produced batch. All of these quality control analysis results need to be within defined limits, to release a batch to the market. This quality control process is very time-consuming mainly due to the laboratory-centered analysis. As part of the doctoral thesis, we have investigated whether historical production and laboratory data for selected medicine could be used for quality predictions of future batches. We have acquired the data for 1005 industry-scale batches. The data for each batch included the results of 53 different laboratory tests of incoming raw materials, intermediates, and final products. Additionally, identification spectra of incoming raw materials and process time series were collected. The data were acquired from different databases, suitably cleaned, and structured for use in prediction models. Identification spectra of incoming raw materials have been successfully used for the prediction of raw materials' quality. Process time series were used for building models that can optimize key process steps. And finally, laboratory data were included in models for the prediction of certain quality attributes of the final product. We have demonstrated that by using only historical data collected and kept by the industry for every single manufactured batch, we can build reliable prediction models which can replace current laboratory analysis. Implementation of models presented in this thesis would save a lot of time, reduce required laboratory capacities and enable the delivery of medicine to patients in a much shorter time.
- Published
- 2023