Classification of drug use patterns

Abstract

Characterizing long‐term prescription data is challenging due to the time‐varying nature of drug use. Conventional approaches summarize time‐varying data into categorical variables based on simple measures, such as cumulative dose, while ignoring patterns of use. The loss of information can lead to misclassification and biased estimates of the exposure‐outcome association. We introduce a classification method to characterize longitudinal prescription data with an unsupervised machine learning algorithm. We used administrative databases covering virtually all 1.3 million residents of Manitoba and explicitly designed features to describe the average dose, proportion of days covered (PDC), dose change, and dose variability, and clustered the resulting feature space using K‐means clustering. We applied this method to metformin use in diabetes patients. We identified 27,786 metformin users and showed that the feature distributions of their metformin use are stable for varying the lengths of follow‐up and that these distributions have clear interpretations. We found six distinct metformin user groups: patients with intermittent use, decreasing dose, increasing dose, high dose, and two medium dose groups (one with stable dose and one with highly variable use). Patients in the varying and decreasing dose groups had a higher chance of progression of diabetes than other patients. The method presented in this paper allows for characterization of drug use into distinct and clinically relevant groups in a way that cannot be obtained from merely classifying use by quantiles of overall use.

Analytical methods for identifying sequences of utilization in health data: a scoping review

“…Classification according to the International Statistical Classification of Diseases and Related Health Problems 10th Revision Chapter Description of Chapter according to ICD-10 classification N (%) Ref. 1 Certain infectious and parasitic diseases 2 (4 %) [ 40 , 50 ] 2 Neoplasms 8 (16 %) [ 7 , 20 , 28 , 32 , 37 , 41 , 44 , 46 ] 4 Endocrine, nutritional and metabolic diseases 10 (20 %) [ 22 , 23 , 29 , 30 , 47 , 49 , 55 , 57 , 59 , 60 ] 5 Mental, Behavioral and Neurodevelopmental disorders 2 (4 %) [ 27 , 35 ] 6 Diseases of the nervous system 2 (4 %) [ 6 , 31 ] 9 Diseases of the circulatory system 10 (20 %) [ 8 , 10 , 24 , 26 , 38 ...…”

“…Further studies use prescription data (11, 22 %) as care events, which can include various events related to medication and treatment. In nearly half of the included records (23, 45 %), visits to medical institutions were analyzed, including consultations in general [ 5 , 45 ] as well as inpatient visits [ 6 , 36 , 46 , 48 , 52 , 53 , 61 ], outpatient consultations [ 6 , 36 , 45 , 48 , 55 , 59 ], hospitalizations [ 30 , 31 , 50 ] and the consultation of specialist physicians [ 5 , 26 , 58 ]. The studies included sometimes considered additional aspects of health care.…”

“…In particular, these studies aimed either to find a group of patients with similar treatment patterns and characterize a population most at risk from a progressing disease (e.g. mental health) [ 27 ], to classify patient groups associated with utilization patterns [ 33 ], to find groups of patients with diabetes who had similar drug use (metformin use) [ 55 ], or to identify therapy profiles [ 51 ]. As noted earlier, the categorization of targets is not unique, and the results of studies looking for trajectories, patterns, or groups of patients with matching characteristics will sometimes have very similar results.…”

“…Objects are assigned initially to cluster centers and reassigned iteratively to minimize an objective function. Partitioning methods were applied less frequently in the identified literature (5, 10 %) [ 21 , 44 , 46 , 55 , 58 ]. In particular, four studies [ 21 , 44 , 46 , 55 ] stated that the k-means algorithm was used.…”

“…Partitioning methods were applied less frequently in the identified literature (5, 10 %) [ 21 , 44 , 46 , 55 , 58 ]. In particular, four studies [ 21 , 44 , 46 , 55 ] stated that the k-means algorithm was used. This was occasionally supplemented by considering specific centres (bayrcentres) [ 46 ] or by conducting several iterations of k-means using ensemble clustering [ 21 ].…”