Search strategy and selection criteria
We systematically reviewed all literature published from 1990 to 2010 to identify studies reporting measures of diarrhea duration and severity in children and adults. From May 20-27, 2010, we searched in Pubmed using combinations of MeSH search terms: diarrhea, gastroenteritis, duration, persistence, severity, infant, child, teenage, and adult.
We initially screened all unique publications for eligibility based on the relevancy of title and abstract; we then screened the full manuscripts for inclusion and exclusion criteria. We included randomized controlled trials (RCTs), cohort, and observational studies published in any language and conducted in any country. Included studies contained data on diarrhea duration and/or severity in children and/or adults. We excluded studies with unclear methodology and diarrhea recall beyond 2 weeks. We accepted studies defining diarrhea as the passage of ≥ 3 watery stools in a 24-hour period; we also considered mother's report of a change in usual stool consistency or frequency for infants ≤ 11 mo of age, and the diagnosis of diarrhea by a physician or nurse as valid case definitions. We excluded studies describing diarrhea deaths alone since duration and severity outcomes of episodes resulting in death are not generalizable to all diarrhea episodes. We also excluded reports of nosocomial outbreaks, diarrhea due to known chronic or non-infectious illness, and studies that limited inclusion to one etiologic agent. For analytical purposes, we did not include studies that combined outcomes across child/adult age categories or inpatient/outpatient status. Furthermore, for duration outcomes, we included studies on the natural course of diarrhea episodes and excluded those reporting on acute or persistent episodes alone.
We abstracted measures of diarrhea duration and severity for all ages; for presentation purposes we grouped these data into 3 distinct age categories: 0-59 mos, 5-15 yrs, and ≥ 16 years of age. For studies stratifying outcomes by treatment or HIV-status, we only abstracted data on placebo and HIV-negative individuals, respectively.
We abstracted three outcomes describing diarrhea duration: mean or median duration--reported as the number of days per episode of diarrhea, and the proportion of total diarrheal episodes that became persistent (≥ 14 or ≥ 15 days). If the proportion of persistent cases was skewed due to an intentional sampling bias or matching, we did not include the study for this outcome. To ensure capture of the full length of diarrheal episodes, we also excluded studies solely reporting the duration of diarrheal episodes at baseline.
Multiple outcomes were reported as measures of severity. We abstracted the proportion of individuals suffering from diarrhea with mild, moderate, severe, or any dehydration; vomiting; and bloody stools. We classified necessitating ORS or intravenous fluids as any dehydration. For studies assessing dehydration by both physician perception and WHO classification for dehydration, we utilized the latter. Likewise, if possible, we abstracted outcomes assessed by physicians or trained health workers in place of those estimated by mothers or caregivers. Additional severity outcomes included the mean stool volume (g/kg/day) and mean stool frequency (stools/day). We also abstracted the proportion with low (< 40 g/kg/day), medium (40-70 g/kg/day), and high (> 70 g/kg/day) stool volume. For stool frequency, we abstracted the proportion low (1-5 stools/day), medium (6-9 stools/day), and high (≥ 10 stools/day), as well as the proportion with > 5 stools/day and > 3 stools/day.
We did not find any studies able to quantify the proportion of episodes that remain mild/moderate or that progress to severe. We therefore used care-seeking behavior among caregivers of children < 5 years of age as a proxy for diarrhea severity in this age group. We abstracted care-seeking behavior from the Demographic Health Surveys (DHS) [6
]. The DHS reports the number of children ≤ 35 months of age with diarrhea in the two weeks preceding the survey for whom care was sought, as well as the number who were not taken to a health provider.
We used STATA 10.1 to generate combined estimates and 95% confidence intervals for all duration and severity outcomes [7
]. To ensure that confidence intervals were lower bound by zero for all outcomes and upper bound by one for proportions, we fit a logistic regression model to data reported as proportions and a Poisson regression model to continuous outcomes. All outcomes were weighted by sample size, which consisted of the total number of episodes. For study designs assessing only one episode per individual, the number of children or adults with diarrhea was considered the total number of episodes. We used the reported diarrheal incidence (number of episodes per child-year) to estimate the sample size for studies evaluating multiple episodes per child without reporting the total number of episodes. We conducted separate analyses for each study type--community, hospital-based inpatient and hospital-based outpatient.
Using the DHS data, we calculated region-specific averages for diarrhea care-seeking. We also calculated the median global estimate of diarrhea care-seeking across regions. The global estimate was used to inform our model describing the total envelope of diarrhea among children under five. Mild cases were defined as those for which no care was sought, and moderate cases were defined as those for which care was sought through a health provider. We assumed that mild and moderate episodes were best described by the duration and severity outcomes derived from community-based studies and hospital outpatient studies, respectively. We used the proportion of outpatient cases presenting with severe dehydration to determine the percentage of moderate cases which progress to severe. Duration and severity outcomes derived from hospital inpatient studies were assumed to best describe severe cases. Table summarizes the algorithms used to determine the proportion of mild, moderate and severe cases occurring among children 0-59 mos of age; Table also lists the source of duration and severity outcomes assumed to best describe each diarrhea category.
Algorithms used to determine the total envelope of childhood and adult diarrhea
We also designed a model to describe the total envelope of diarrhea among individuals ≥ 16 years of age. We assumed that mild cases were comprised of cases occurring in the community without dehydration, and moderate cases were comprised of the proportion of community cases with any dehydration. In keeping with the assumptions made for the model describing the total envelope of diarrhea among children under-five, we used the proportion of outpatient cases presenting with severe dehydration to determine the percentage of moderate cases progressing to severe. We also assumed that community, hospital outpatient, and hospital inpatient studies best described the duration and severity profiles of mild, moderate, and severe cases, respectively (See Table ).