Abstract
BACKGROUND: Evaluation of parasite clearance patterns in experimental human infection trials helps increase understanding of drug action. In a previously reported phase Ib trial of a new investigational anti-malarial drug M5717, parasite clearance showed a biphasic linear pattern: slow removal phase with a near flat clearance rate followed by a fast clearance phase with a steep slope. In this study three statistical approaches were implemented and compared to estimate the parasite clearance rate for each phase and the time point corresponding to the change of clearance rates (changepoint between the two phases). METHODS: Data using three M5717 doses 150 mg (n = 6), 400 mg (n = 8), 800 mg (n = 8) were used to estimate biphasic clearance rates. Three models were investigated: firstly, segmented mixed models with estimated changepoint-models with/without random effects in various parameters were compared. Secondly, a segmented mixed model using grid search-this method is similar to the first except that changepoints were not estimated, instead they were selected based on model fit from given candidate values. Thirdly, a two-stage approach whereby a segmented regression model fit to each participant followed by a meta-analysis method. Hourly rate of parasite clearance (HRPC) interpreted as the percentage of parasites removed each hour was calculated. RESULTS: The three models generated similar results. Using segmented mixed models, the estimated changepoints after treatment in hours (95% CI) were: 150 mg: 33.9 (28.7, 39.1); 400 mg: 57.4 (52.5, 62.4); and 800 mg: 52.8 (47.4, 58.1). For all three treatment groups, there was nearly no clearance before the changepoints, but rapid clearance in the second phase (HRPC [95% CI]): 150 mg: 16.8% (14.3, 19.1%); 400 mg: 18.6% (16.0, 21.1%); and 800 mg: 11.7% (9.3, 14.1%). CONCLUSIONS: All three statistical approaches are effective tools to characterize the bi-phasic clearance of M5717 in the phase 1b experimental Plasmodium falciparum malaria human infection study. The statistical approaches produced similar results to estimate the two-phase clearance rates and the changepoint for each treatment dose of M5717. However, the segmented mixed model with random changepoints has several advantages: it is computationally efficient, provides precision for changepoint estimates and is robust concerning outlying datapoints or individuals.