Quinten solutions - Advisory and evidence generation services

Treatment patterns mining

Machine learning algorithms developed by Quinten Health allow for characterizing optimal treatment patterns based on patient’s medical history and characteristics

Using dedicated sequence pattern mining technologies, switching patterns and disease progression over time are profiled using a combination of interpretable classifiers or predictors. Such approaches allow for personalized care decisions and disease-centric evaluations.

Use case examples include:


  • Detection of unmet medical needs in the real-world practice
  • Public health impact assessment
  • Optimal drug positioning
  • Treatment pathway optimization and medical practice guidance
  • Drug utilization monitoring and projection in real-world practice

Treatment mining for a new drug positioning

Context: understanding the existing medical practice to define new drug positioning

Numerous concrete medical questions can be tackled via treatment patterns mining. What is the best time to treatment initiation? How to adapt treatment strategies to patients’ profiles? What are the most frequent sequences of medications? What are the clinical characteristics of patients responding to these sequences or refractory to them? What is the overall population long-term benefit for a new drug in the treatment pathway? As an illustrative case study, Quinten Health had to optimize positioning of a new drug within the care pathway of patients with a chronic respiratory disease.

Quinten Health’s solution: sequence mining algorithms

Using longitudinal data from claims databases and registries, Quinten Health adapted and compared state of the art sequence mining algorithms [1], [2] to uncover the most frequent treatment pathways. The treatment sequence profiles were then mapped to sequences of clinical outcomes (from disease aggravation or relapse to disease control or full recovery) accounting for the patient condition.

Impact: fact-based decision making

As a result, the most effective treatment sequences were evaluated given a patient profile. Typical results were first delivered in the form: “prescribing drug C in non-responders to drug A occurred in 33% of patients with controlled disease and in only 13% of patients with a relapse”. Then for each typical patient profile, optimal treatment pattern was displayed, together with a comparative benefit vs. standard of care and prevalence/incidence of each profile. Embedding such technology in a visualization dashboard to facilitate interactions shown to be impactful in guiding patient-centric medical practice and public health evaluations.


[1] Zaki, M. J. SPADE: An efficient algorithm for mining frequent sequences. Machine learning, 2001, vol. 42, no 1-2, p. 31-60.

[2] Guo, C., Chen, J. Big Data Analytics in Healthcare: Data-Driven Methods for Typical Treatment Pattern Mining. J. Syst. Sci. Syst. Eng., 2019, 28, p. 694–714

See our other cases