The statistical analysis plan [53 ] was developed before analysis. All non-synonymous mutations in the pfk13 gene identified in the studies were included in the analysis. Isolates without reported mutations were assumed to be wild type in assessing relationships between parasite genotype and PC1/2. Isolates with a mixed genotype at any nucleotide within the pfk13 coding region (wild type/mutation or two non-synonymous mutations) were excluded from the analysis.
The PC1/2 is defined as the time in hours needed for the parasite density to decline by 50% during the log-linear phase of parasite clearance. PC1/2 was calculated using the WWARN parasite clearance estimator tool [41 (link)]. The goodness of fit of parasite clearance models was evaluated for each individual patient parasitemia-time profile used to estimate the PC1/2.
Profiles that satisfied the following criteria (i.e., provided biologically or statistically plausible results) were included in the analysis: (a) standard deviations of residuals < 2, (b) number of data points used to fit the linear part of the curve > 2, (c) duration of lag phase < 12 h, (d) pseudo R2 statistics ≥ 0.8. Additionally, patients who withdrew or had a record of inadequate dosing were excluded. The log transformed half-life metric was modeled for all pfk13 mutant alleles in all studies with information from individual patients on age, initial parasitemia, ACT treatment, and artesunate dose as covariates. The method by which the dose was calculated is documented in the statistical analysis plan. Random effects for study site were used to account for heterogeneity between studies. Residuals were examined for normality and for systematic deviations from the model.
The differences in PC1/2 between infections with P. falciparum parasites bearing a specific pfk13 propeller mutant allele and those with wild type parasites were assessed by the Wald test. The fold change in geometric mean of PC1/2 of infections with pfk13 mutant parasites compared to wild type isolates from the same sites; xPC1/2 was calculated as an exponent of the difference of the corresponding regression coefficients.
In order to determine a PC1/2 threshold value that defined slow parasite clearance, we divided Asian data into two populations: rapid clearing and slow clearing. The slow-clearing population was defined as all isolates with mutations associated with a significant increase in PC1/2 values in this analysis, while the fast-clearing population included all other isolates. The PC1/2 value that corresponds to the 95th percentile of the fast-clearing population (i.e., a value x such that the probability that PC1/2 > x is less than 0.05) was selected as the cutoff for infections with “slow clearing” parasites. Risk of bias in individual studies was assessed based on frequency of parasite counting, molecular methodology, and number of patients excluded because of missing data or unsatisfactory fit of the model for PC1/2 estimation (for details, see Additional file1 ). Data from studies/sites that reported results very different from all of the others in the same region were included in the analysis, and a sensitivity analysis was conducted after excluding Tra Lang, Vietnam (study site ID 23; study ID 8), and Pyin Oo Lwin, Myanmar (study site ID 15; study ID 13).
The PC1/2 is defined as the time in hours needed for the parasite density to decline by 50% during the log-linear phase of parasite clearance. PC1/2 was calculated using the WWARN parasite clearance estimator tool [41 (link)]. The goodness of fit of parasite clearance models was evaluated for each individual patient parasitemia-time profile used to estimate the PC1/2.
Profiles that satisfied the following criteria (i.e., provided biologically or statistically plausible results) were included in the analysis: (a) standard deviations of residuals < 2, (b) number of data points used to fit the linear part of the curve > 2, (c) duration of lag phase < 12 h, (d) pseudo R2 statistics ≥ 0.8. Additionally, patients who withdrew or had a record of inadequate dosing were excluded. The log transformed half-life metric was modeled for all pfk13 mutant alleles in all studies with information from individual patients on age, initial parasitemia, ACT treatment, and artesunate dose as covariates. The method by which the dose was calculated is documented in the statistical analysis plan. Random effects for study site were used to account for heterogeneity between studies. Residuals were examined for normality and for systematic deviations from the model.
The differences in PC1/2 between infections with P. falciparum parasites bearing a specific pfk13 propeller mutant allele and those with wild type parasites were assessed by the Wald test. The fold change in geometric mean of PC1/2 of infections with pfk13 mutant parasites compared to wild type isolates from the same sites; xPC1/2 was calculated as an exponent of the difference of the corresponding regression coefficients.
In order to determine a PC1/2 threshold value that defined slow parasite clearance, we divided Asian data into two populations: rapid clearing and slow clearing. The slow-clearing population was defined as all isolates with mutations associated with a significant increase in PC1/2 values in this analysis, while the fast-clearing population included all other isolates. The PC1/2 value that corresponds to the 95th percentile of the fast-clearing population (i.e., a value x such that the probability that PC1/2 > x is less than 0.05) was selected as the cutoff for infections with “slow clearing” parasites. Risk of bias in individual studies was assessed based on frequency of parasite counting, molecular methodology, and number of patients excluded because of missing data or unsatisfactory fit of the model for PC1/2 estimation (for details, see Additional file
Full text: Click here