|
|
 |
| |
|
|
|
A Random Survival Forest Model for Time-to-Virologic Failure and Drug Resistance: Randomized Factorial Design using Data from Botswana |
|
|
|
PP: 1355-1370 |
|
|
doi:10.18576/amis/200518
|
|
|
|
Author(s) |
|
|
|
Claris Shoko,
Kesaobaka Molebatsi,
Katleho Makatjane,
Ontlametse Thato Choga,
Sikhulile Moyo,
|
|
|
|
Abstract |
|
|
| We analyse the clinical and drug-related factors that predict time to treatment failure and development of drug resistance among people living with HIV (PLWH). We demonstrate this on an open-label, randomized, 3 × 2 × 2 factorial design study that was conducted in Botswana from 2002 to 2007, involving 650 antiretroviral therapy (ART)-na ̈ıve individuals. To identify predictors associated with virologic failure and drug resistance, variables were initially pre-screened using the log-rank test. A Random Survival Forest (RSF) model was built with the log-rank splitting rule and 80 trees. The model’s performance was assessed using Brier scores, out-of-bag error, concordance index (C-index) and continuous ranked probability scores (CRPS). To further examine variable importance and interaction effects, paired variable significance measures and maximum subtree depth were also analyzed. The RSF model showed excellent prediction accuracy, with ensemble survival probability over 90% and low Brier scores across all risk percentiles. The key predictors of virologic failure were the non-nucleoside transcriptase inhibitors (NRTI) class, treatment arm, baseline CD4 count of 250 cells/mm3 or less, World Health Organisation (WHO) HIV/AIDS clinical stage 4 (severe symptoms), severe anemia, and wasting syndrome. While Stavudine/Lamivudine (D4T/3TC)-based regimens were associated with a low risk of virologic failure and drug resistance, Zidovudine/Didanosine (ZDV/ddI)-containing regimens were associated with a higher risk of virologic failure and drug resistance. We found a significant interaction between drug regimens and clinical stage. The probabilities of survival from virologic failure with resistance increased with time on treatment. Random Survival Forest provides a robust and flexible tool for survival analysis in HIV research, showing significant handling complex interactions. The findings support risk-based patient monitoring and regimen optimization to enhance virologic suppression for people living with HIV (PLWH)..
|
|
|
|
|
|
|
|
