Our study demonstrated recall decay of symptoms to be a robust phenomenon occurring at multiple levels—for both children and adults, for three key symptoms, for self-reported and proxy-reported symptoms in adults, for severe and non-severe illness in children and in rural and urban African settings. This finding suggests that rates of diseases identified in different studies might vary as a result of not only differing demographics, environmental conditions and case definitions, but also differences in the recall period used.3,4
Previous studies on recall decay focused on children in single study sites and all but one on single syndromes.15–19
In Bangladesh, there was a 34% decrease in diarrhoea episodes among toddlers reported 48 h prior to the interview time.15
In India, the proportion of infants in whom diarrhoea was reported was similar up until day 3 prior to the interview date, thereafter showing a steady decline with a 15, 26 and 45% decrease in the proportion of children with diarrhoea reported on days 3, 6 and 7–13, respectively, before the interview date.19
Under-reporting of diarrhoea episodes has also been shown to occur if the recall period is >2–3 days in DHS surveys that ask about diarrhoea in the past 2 weeks.16
A study from the Gambia that carried out weekly visits showed a ~50% decrease in the symptoms of diarrhoea, fever and respiratory illness among infants between the day of the visit and 8 days before.17
Another study of maternal recall of symptoms of acute respiratory illness showed that the specificity of the symptoms recalled was better 2 weeks after the diagnosis than 4 weeks after.18
The findings of our study have implications for designing studies to define the rate of disease using surveillance methods based on recall of symptoms. The period of recall should probably not extend back >3 days for children and >4 days for adults to achieve a rate that is at least 80% of the possible maximum rate. To achieve complete ascertainment of symptoms during longer time periods, more frequent visits than fortnightly or weekly visits should be considered. More frequent household visits have been made in some studies, but are resource intensive and likely to lead to participant fatigue over time.3,11,20–22
An alternative strategy to continuous morbidity surveillance for defining disease rates is to record symptoms in only a fraction of days.23,24
Sampling at 7- or 14-day intervals, asking about symptoms 24–48 h before the visit, was shown to result in a relatively small loss of precision when estimating longitudinal prevalence of diarrhoeal disease, requiring only small increases in sample size.24
Our data suggest that the precision of such intermittent sampling strategies might be increased by using 3–4-day recall periods without compromising accuracy too much. However, the interval of sampling was sensitive to the duration of the disease episodes, and so intermittent sampling might apply more to measurements of longitudinal prevalence than incidence.23,24
If periods of recall >3–4 days are considered, mnemonics such as symptom diaries might be employed to minimize recall decay, although the compliance and accuracy with such devices in populations with low literacy have been questioned.17,25
To our knowledge, this is the first report of recall decay for clinic visitation and drug use. This finding has implications for surveys assessing health-seeking patterns and drug use in communities, such as the DHSs, which ask about clinic use and medications in the past 2 weeks. If a recall period of >1 week is used, clinic utilization and drug use will likely be underestimated. Recall of antibiotic use was lower than anti-malarial use likely because anti-malarials are more recognizable by name than are antibiotics in rural western Kenya.
Our study evaluated the impact of several different factors on symptom recall that shed light on how human memory of symptoms works. First, we showed that proxy-reported symptoms have steeper recall decay than self-reported symptoms, which would be expected since individuals remember their own experiences better than other people do.26
Secondly, we found that people in the urban setting seemed to remember their symptoms better than people in the rural setting. The reason for this is not clear, but it could be related to the urban population being more educated and aware of dates due to salaried employment, compared with the rural setting where most persons are subsistence farmers. Alternatively, because rates of reported disease are higher in the rural setting, recall of separate episodes of similar illness might be more difficult, a phenomenon that has been shown.26
Our study did not support the so-called ‘salient principle’, whereby more severe symptoms are remembered longer than milder symptoms, which has been observed before, although the finding is not consistent across studies.5,17
Lastly, our data show that recall seemed to drop precipitously 1 week before home visit, suggesting that recall of symptoms might be packaged into week blocks, rather than in days.
Symptom recall by participants might have been biased by several aspects of human memory.26
First, ‘telescoping’ of memory, in which events are remembered to have occurred more recently than they actually did, could have occurred.26,27
It is possible that symptoms that occurred 7–13 days before the home visit might have been recalled as having occurred during days 0–6 before home visit, thereby falsely elevating the rates calculated for the last week. This phenomenon of telescoping has been documented with events occurring over longer periods of recall of months or years and might not tend to occur as markedly in a 2-week recall period, particularly during the past 3–4 days. Secondly, people tend not to remember dates as well as events in an autobiographical sequence.26
Therefore, by asking participants to remember symptoms in relation to a personal event, such as a market day or a church day, rather than in a 2-week period, might have led to greater accuracy of recall, particularly for less recent symptoms. Lastly, participants might have tended to over-report symptoms because they wanted to be ‘helpful’ interviewees, whereas on the other hand they might have under-reported symptoms due to interview fatigue during longitudinal surveillance.26
Although both of these occurrences might have yielded inaccurate rates of disease, they would likely not have affected this analysis of recall as they would have been unlikely to have been differential with respect to symptom recall period.
In conclusion, because of under-utilization of health facilities, defining disease burden in developing country settings can be more complete when asking persons to recall symptoms of recent illnesses during home visits. Recall of symptoms is greatly influenced by the time period of recall. Our data suggest that the period of recall should not exceed 3 days in children and 4 days in adults to identify at least 80% of the maximum rate of disease, particularly for longitudinal studies involving repetitive visits. Limitations of symptom recall should be considered in the context of the logistics and cost of designing a surveillance system to define the burden of disease in developing countries.