All efficacy studies begin with patients enrolled in a study protocol and treated for malaria. Most studies end when the patient meets the criteria for failure or the patient is followed for some pre-determined duration without meeting criteria for failure. Given variations in duration of follow-up, survival analysis is generally regarded as the most appropriate analytical approach [12
]. If a patient observation period ends prematurely and treatment failure did not occur, then that patient is "censored" at the time s/he was last observed.
The use of survival analysis offers the following important advantages :
1) Data from patients with different follow-up periods can be combined and efficacy estimates generated at different time points. These can then be compared between studies with different length of follow-up.
2) All available data contributes in the analysis, thus increasing the precision of the derived estimates.
3) The approach avoids systematic biases introduced by dropping patients from the analysis that do not complete follow-up (per-protocol analysis) or classifying patients as failures who do not represent true biological failures (intention-to-treat analysis).
In survival data each patient is characterized by a period of observation and whether or not treatment failure occurred during this observation period (the "status"). This can be derived from four key parameters:
1) The last day of follow-up
2) The patient outcome on the last day of follow-up
3) The parasite species on the day of failure
4) The genotyping results (recrudescence, reinfection, no result)
The outcomes for all patients on the last day of follow-up need to be categorized into mutually exclusive groups that cover all possible endpoints (Table ). These can be broadly divided into three groups.
Key variables requested by the global efficacy database
1) Patients who complete the study and do not meet criteria for treatment failure. For these patients, the last day of follow-up is the full duration of the study period.
2) Patients who meet criteria for treatment failure; in this case the last day of follow-up is the day when treatment failure occurred.
3) Patients who neither complete the full follow-up period nor meet the criteria for treatment failure (e.g. lost to follow-up, withdrawal of consent, use of other antimalarials, protocol violations, infection with other species etc.). For these patients, the last day of follow is the day when the patient was last observed.
Since the primary objective is to determine the prevalence of parasite resistance, failure is defined as early treatment failure during the first few days after the start of treatment or the first reappearance or persistence of parasitaemia during subsequent follow-up. These definitions are consistent with the current WHO guidelines.
Condensing the patients' in vivo response into the four simple parameters listed above, allows one to generate in a transparent and flexible manner, the status of the patient required for survival analysis. Table shows an example of a coding table which allows reviewers to appreciate how various protocol violations were dealt with and thus the shortcomings of the derived estimates. If additional baseline data are available, one can then stratify the efficacy estimates according to important baseline line confounding variables (Table ).
Description of possible outcomes on the last day of follow-up for patients enrolled in clinical efficacy studies
The derivation of the patients status on the last day of follow, which is required for survival analysis
The proposed system standardizes the process of data collection in a robust and yet flexible manner, allowing an analysis that accommodates the inevitable diversity of study methodologies and collection methods. Importantly it still retains the ability to present alternative analyses, such as proportions of failures on any particular day. Whatever analytical method is used the process should be transparent to analysts interpreting the findings. Such transparency is now evident from the coding table (Table ). This approach of using standardized terminology greatly facilitates the ability to upload and cross check individual patient study data to a global resource. Once pooled estimates of efficacy can be readily generated and compared across geographically and temporally disparate studies.
Data can be gathered retrospectively from completed studies but this may require recoding of the data to produce the key parameters outlined in Table (for further details see reference [13
]. If the collection of these key parameters are incorporated prospectively into studies, then the process is simplified greatly.