Artificial intelligence algorithms for dialysis patients' therapies efficiency evaluation
The aim of the research is to form methodological base of medical information system development, that would be able to examine the life quality problems of dialysis patients as a whole by using machine learning algorithms.
A specific practical objective is to develop an intelligent decision support system for prescribing personalized medical therapies for patients with chronic renal failure, as well as to evaluate the efficiency of the treatment strategy in terms of the validity of prescribing for phosphorus-calcium metabolism (PCM) restoration therapy and for antianemic therapy (AAT) based on patient's profile. By patient's profile we understand the combination of socio-demographic characteristics of the patient, functional examinations, laboratory and clinical studies, monitored in dynamics, the timeline of pharmacological prescriptions in the "drug-dose-route of administration" link.
Material and methods. The model of therapies efficiency had been developed using the funnel principle: on the first stage of the model therapy classifies as "effective" one or "non-effective", then in the case of "non-effective" therapy classifies either as "insufficient" or as "excessive". In the research algorithms of gradient boosting and random forest were implemented on both stages. To balance volume of raw data recovery and to get reliable results while having a wide variety of features sampling technics as SMOTE and random oversampling were used. As features for classification models fitting were used such values as: gender and age of patient, body mass index, presence of hepatitis B, hepatitis C, HIV infection, treatment period, presence or absence of medical treatment prescription in previous periods with medicament indication, its dosage, medication frequency and route of administration, laboratory blood indicators at the moment of therapy prescription (such as hemoglobin, ferritin, potassium, sodium, hematocrit, phosphorus, iron, parathormone, calcium, percentage of transferrin iron saturation, etc.), their values at previous and pre previous months, duration of dialysis treatment, dialysis procedure efficiency indicator (monthly average KT/V ratio). The data that was used to fit models is represented by 9000 records labeled by efficiency class for both AAT and for PCM restoration therapy. Given data was divided on test and train data in 70 on 30 ratio.
Results. On test data the following quality metrics were received for therapy efficiency estimation using fitted models: for AAT efficiency - sensitivity -98.9%, specificity - 98.2%; for PCM restoration efficiency - sensitivity - 98.4%, specificity - 98.3%; for AAT insufficiency/excessiveness - sensitivity - 98.4%, specificity - 97.7%; for PCM restoration therapy insufficiency/excessiveness -sensitivity - 99.5%, specificity - 100%.
Conclusion. The implementation of the proposed algorithms efficiency estimation for AAT and PCM recovery therapy system allows to use given means for these types of therapy as efficiently as possible.
Funding. The work, the results of which are presented in the article, was partially supported by the grant “Development of an intelligent decision support system for the appointment of personalized dialysis and drug therapy for patients with chronic renal failure using artificial intelligence algorithms” (grant of the Innovation Research and Development Assistance Fund, 2019-2020, No. AAAAA-A20-120011490126-5).
Conflict of interest. The authors declare no conflict of interest.
Contribution. Statement of the research problem - Chernenko O.V.; development of intelligent algorithms -Lakman L.A.; study design - Shkel O.A.; assessment of the predictive quality of intelligent algorithms - Padu-kova A.A.; collection and primary analysis of data - Nafikov Sh.R.; software development for intelligent algorithms - Shabanova K.I.
For citation: Chernenko O.V., Lakman L.A., Shkel O.A., Padukova A.A., Nafikov Sh.R., Shabanova K.I. Artificial intelligence algorithms for dialysis patients' therapies efficiency evaluation. ORGZDRAV: novosti, mneniya, obuchenie. Vestnik VSHOUZ [HEALTHCARE MANAGEMENT: News, Views, Education. Bulletin of VSHOUZ]. 2021; 7 (2): 103-15. DOI: https://doi.org/10.33029/2411-8621-2021-7-2-103-115 (in Russian)
1. Andrusev A.M., Peregudova N.G., Shinkarev M.B., Tomilina N.A. Replacement therapy for end-stage chronic renal failure in the Russian Federation in 2014–2018. A brief report on the data of the Russian Register of Renal Replacement Therapy. 19 p. URL: http://www.nephro.ru/content/fi les/registr/Registr_2014-2018_short.PDF (in Russian)
2. Usvyat L., Dalrymple L.S., Maddux F.W. Using technology to inform and deliver precise personalized care to patients with end-stage kidney disease. Semin Nephrol. 2018; 38 (4): 418-25. DOI: https://doi.org/10.1016/j.sem-nephrol.2018.05.011
3. Hueso M., Vellido A. Artificial intelligence and dialysis. Kidney Dis. 2019; 5: 1-2. DOI: https://doi.org/10.1159/000493933
4. Von Gersdorff G.D., Usvyat L., Marcelli D., Grass-mann A., Marelli C., Etter M., et al. Monitoring dialysis outcomes across the world - the MONDO global database consortium. Blood Purif. 2014; 36 (3-4): 165-72. DOI: https://doi.org/10.1159/000356088
5. Hueso M., Vellido A., Montero N., Barbieri C., Ramos R., Angoso M., et al. Artificial intelligence for the artificial kidney: pointers to the future of a personalized hemodialysis therapy. Kidney Dis. 2018; 4: 1-9. DOI: https://doi.org/10.1159/000486394
6. Lankhorst C.E., Wish J.B. Anemia in renal disease: diagnosis and management. Blood Rev. 2010; 24: 39-47. DOI: https://doi.org/10.1016/j.blre.2009.09.001
7. Brier M.E., Gaweda A.E. Artificial intelligence for optimal anemia management in end-stage renal disease. Kidney Int. 2016; 90: 259-61. DOI: https://doi.org/10.1016/j.kint.2016.05.018
8. Martmez-Martmez J.M., Escandell-Montero P., Barbieri C., Soria-Olivas E., Mari F., Martinez Sober M., et al. Prediction of the hemoglobin level in hemodialysis patients using machine learning techniques. Comput Methods Programs Biomed. 2014; 117: 208-17. DOI: https://doi.org/10.1016/j.cmpb.2014.07.001
9. Zinovev D.A., Novitskiy V.O., Malkoch A.V. Application of a neural network machine learning method to complication diseases treatment in hemodialysis patients. Vrach i in-formatsionnye tekhnologii [Doctor and Information Technologies]. 2019; (2): 29-37. (in Russian)
10. Fukagawa M., Kido R., Komaba H., Onishi Y., Ya-maguchi T., Hasegawa T., et al. Abnormal mineral metabolism and mortality in hemodialysis patients with secondary hyperparathyroidism: evidence from marginal structural models used to adjust for time-dependent confounding. Am J Kidney Diseases. 2014; 63 (6): 979-87. DOI: https://doi.org/10.1053/j.ajkd.2013.08.011
11. Chawla N., Bowyer K., Hall L., Kegelmeyer W. SMOTE: synthetic minority over-sampling technique. J Artif Intell Res. 2002; 16: 341-78. DOI: https://doi.org/10.1613/jair.953
12. Cangelosi R., Goriely A. Component retention in principal component analysis with application to cDNA microarray data. Biol Direct. 2007; 2: 2. DOI: https://doi.org/10.1186/1745-6150-2-2
13. Witten I.H., Frank E. Data Mining: Practical Machine Learning Tools and Techniques. 2nd ed. Morgan Kaufmann, 2005: 558 p.
14. Airola A., Pahikkala T., Waegeman W., De Baets B., Salakoski T. An experimental comparison of cross-validation techniques for estimating the area under the ROC curve. Comput Stat Data Anal. 2011; 55 (4): 1828-44. DOI: https://doi.org/10.1016/j.csda.2010.11.018
15. Escandell-Montero P., Chermisi M., Martmez-Mar-tfnez J.M., Gomez-Sanchis J., Barbieri C., Soria-Olivas E., et al. Optimization of anemia treatment in hemodialysis patients via reinforcement learning. Artif Intell Med. 2014; 62 (1): 47-60. DOI: https://doi.org/10.1016/j.artmed.2014.07.004
16. Press W.H., Flannery B.P., Teukolsky S.A., Vetter-ling W.T. Singular value decomposition. § 2.6. In: Numerical Recipes in FORTRAN: The Art of Scientific Computing. 2nd ed. Cambridge, England: Cambridge University Press, 1992: 51-63.