OBJECTIVES.

The study purpose was to describe the changes and discontinuations of antiretroviral combination therapy (cART) (number of treatments, occurrence, duration) over the cohort of incident people living with HIV (PLWHIV) in 2013, included in the French national Health Insurance database (SNIIRAM). It also aims at experimenting data mining methods to study treatment sequences.

METHODS.

From the SNIIRAM, we extracted PLWHIV affiliated to the general insurance scheme (N=96,423) through specific chronic diseases status and/or reimbursement of HIV laboratory tests and/or HIV-related hospitalizations and/or reimbursements of antiretroviral drugs in 2013. Incident patients (N=3,373) were followed for 2 years to identify their cART, categorized as single, double, triple or quadruple+ combination therapies. The treatment sequences were arranged according to a similarity criterion. It was performed by an Agglomerative Clustering algorithm configured with the Hamming distance and the Ward linkage method. The result is a graph containing one discretized timeline per patient. The lines were put in order one above the other and follow the 2-year follow-up. Finally, a smoothing was applied to the image to homogenize and highlight distinctive patterns. The combination of Agglomerative Clustering and of image processing forms an innovative methodology called TAK: Time-sequence Analysis through K-clustering.

RESULTS.

Over the 3,373 incident patients, 363 (11%) received no antivirals drugs during the follow-up, 385 (11%) were only treated after 6 months and 255 (8%) stopped cART at least one period during follow-up. 1,675 (50%) patients solely received a triple therapy over the follow-up, 484 (14%) changed at least once from a triple therapy to a single/double.

CONCLUSIONS.

In 2013, a triple therapy was the reference treatment and 30% of the patients had a 6-month period without any cART. This study validates the relevance of combining clustering methods with image processing to visualize the treatment sequences of a cohort in a clear and synthetic way.