Recall bias in epidemiologic studies

The factors which contribute to bias due to differential recall between cases and controls in retrospective studies have been little studied. A review of the literature on recall accuracy suggests that the extent of inaccurate recall is related to characteristics of the exposure of interest and of the respondents, though a distinction must be drawn between recall which is biased and that which is simply inaccurate. Interviewing technique and the study protocol, including the design of questionnaires and the motivation of respondents, play a central role and are under the control of the investigator. The results of validation studies carried out to date suggest that the likelihood of recall bias may be greater when recall is poor in general.     

Evidence of recall bias in volunteered vs. prompted responses about occupational exposures

BACKGROUND: Recall bias remains a concern in case-control studies, although few investigations have found evidence of differential recall. This study examined whether differences in occupational exposure reporting occur in volunteered vs. prompted questionnaire responses. METHODS: In a large, population-based, case-control study of a childhood cancer, neuroblastoma, we calculated odds ratios for broad occupational exposure groups on the assumption that in the absence of recall bias, risk estimates for such broad groupings should be close to the null value. RESULTS: Prompted exposures and work activities showed little evidence of differential recall by parents of cases and controls (all OR < 1.2), but case parents were more likely to volunteer information about other exposures or activities (ORs: 1.35-1.71). Case mothers were also more likely than control mothers to report activities involving indirect exposure (OR = 1.41). CONCLUSIONS: These findings suggest that prompted exposure questions are less likely to be subject to recall bias than open-ended questions.     

The effects of recall errors and of selection bias in epidemiologic studies of mobile phone use and cancer risk

This paper examines the effects of systematic and random errors in recall and of selection bias in case-control studies of mobile phone use and cancer. These sensitivity analyses are based on Monte-Carlo computer simulations and were carried out within the INTERPHONE Study, an international collaborative case-control study in 13 countries. Recall error scenarios simulated plausible values of random and systematic, non-differential and differential recall errors in amount of mobile phone use reported by study subjects. Plausible values for the recall error were obtained from validation studies. Selection bias scenarios assumed varying selection probabilities for cases and controls, mobile phone users, and non-users. Where possible these selection probabilities were based on existing information from non-respondents in INTERPHONE. Simulations used exposure distributions based on existing INTERPHONE data and assumed varying levels of the true risk of brain cancer related to mobile phone use. Results suggest that random recall errors of plausible levels can lead to a large underestimation in the risk of brain cancer associated with mobile phone use. Random errors were found to have larger impact than plausible systematic errors. Differential errors in recall had very little additional impact in the presence of large random errors. Selection bias resulting from underselection of unexposed controls led to J-shaped exposure-response patterns, with risk apparently decreasing at low to moderate exposure levels. The present results, in conjunction with those of the validation studies conducted within the INTERPHONE study, will play an important role in the interpretation of existing and future case-control studies of mobile phone use and cancer risk, including the INTERPHONE study.     

Recall bias did not affect perceived magnitude of change in health-related functional status

BACKGROUND AND OBJECTIVE: It was hypothesized that within an invasively treated group and within a group that improved in angina pectoris no difference in effect size would occur between prospective and retrospective measures. Furthermore, it was hypothesized that assessment of perceived change at post-test may be invalid because of recall bias and present-state bias. STUDY DESIGN AND SETTING: Effect sizes (as standardized response means) were used as indicators of magnitude of change. Linear structural equation analysis (with LISREL) was used to investigate the relationship between the estimates of recall accuracy and retrospectively assessed change. RESULTS: No significant differences were found between prospective and retrospective measures of change over time in health-related functional status. Recall bias was not associated with retrospective measurement of change within a 12-week interval. An expected present-state effect was found in a structural equation model. CONCLUSION: Prospective and retrospective indices of magnitude of change were similar between groups receiving treatment of known efficacy. Recall bias seems to be an acceptable risk in short-term follow-up studies.     

Early symptoms of ovarian cancer: a case-control study without recall bias

BACKGROUND: Ovarian cancer is usually diagnosed after it has spread and is difficult to cure. Previous attempts to identify early symptoms have either lacked a control group or have been based on interviews of cases, with possible recall bias. OBJECTIVE: The purpose of this study was to identify early symptoms of ovarian cancer by reviewing prediagnostic medical records, free of recall bias, and comparing women with and without ovarian cancer. METHODS: In an integrated health care delivery system, symptoms recorded in medical records of 102 women with ovarian cancer during the two years before diagnosis were compared with those of 102 matched control women. RESULTS: More cases than controls complained of several symptoms up to one year before diagnosis. Most of these symptoms were abdominal or gastrointestinal in nature and were more prevalent in the advanced stage cases. Other symptom sites included pelvic, urinary, back, and systemic. Because case-control differences were not large and prevalence is low, positive predictive values were generally quite low. CONCLUSION: Previous reports of early symptoms of ovarian cancer were confirmed in a study with a control group and free of recall bias. It is not clear that these symptoms occurred while the disease was still localized. Because hundreds of women would have to be investigated to detect one case of ovarian cancer, the clinical utility of these symptoms is uncertain. Nevertheless, health care providers should keep ovarian cancer in mind, when women present with symptoms such as abdominal pain and bloating.     

Indications of recall bias found in a retrospective study of physical activity and myocardial infarction

OBJECTIVE: To evaluate the presence of recall bias in retrospective studies of physical activity and myocardial infarction. STUDY DESIGN AND SETTING: In 2005, seventy-eight cases who had suffered from a myocardial infarction and 243 control subjects, who had previously participated in the Swedish WOLF (Work, Lipids, and Fibrinogen) study, were asked about their physical activity level during 1990-2005. The answers about recalled past leisure time, occupational, and household physical activity level were compared with physical activity level as reported at the baseline examination of the WOLF study in 1992-1998. RESULTS: The proportion who recalled the same activity level as originally reported ranged from 69% to 96% (cases) and 69% to 89% (controls), and the kappa values ranged from 0.30 to 0.91 (cases) and 0.46 to 0.59 (controls), with the exception of perceived physical workload in household work, which showed low agreement between the originally stated and later recalled activity levels. Some differences were found between cases and controls regarding recall of past occupational activity, indicating the presence of recall bias in this domain of physical activity. CONCLUSION: We cannot preclude the existence of recall bias when using retrospectively recalled information about occupational physical activity in studies of physical activity and myocardial infarction.     

Parr et al. Respond to "Recall Bias in Melanoma"

In this issue of the Journal, our study of recall bias in melanomarisk factors (1) is presented together with an invited commentaryby Gefeller (2). On the basis of our results, Gefeller concludesthat recall bias may not be a reason for great concern afterall. Although we think that our study showed some evidence of     

Accuracy of patient recall for self‐reported doctor visits: Is shorter recall better?

In health economics, the use of patient recall of health care utilisation information is common, including in national health surveys. However, the types and magnitude of measurement error that relate to different recall periods are not well understood. This study assessed the accuracy of recalled doctor visits over 2‐week, 3‐month, and 12‐month periods by comparing self‐report with routine administrative Australian Medicare data. Approximately 5,000 patients enrolled in an Australian study were pseudo‐randomised using birth dates to report visits to a doctor over three separate recall periods. When comparing patient recall with visits recorded in administrative information from Medicare Australia, both bias and variance were minimised for the 12‐month recall period. This may reflect telescoping that occurs with shorter recall periods (participants pulling in important events that fall outside the period). Using shorter recall periods scaled to represent longer periods is likely to bias results. There were associations between recall error and patient characteristics. The impact of recall error is demonstrated with a cost‐effectiveness analysis using costs of doctor visits and a regression example predicting number of doctor visits. The findings have important implications for surveying health service utilisation for use in economic evaluation, econometric analyses, and routine national health surveys.     

Invited Commentary: Recall Bias in Melanoma--Much Ado About Almost Nothing?

Recall bias has been given considerable attention in textbooksand methodological research because of its potential to jeopardizethe validity of epidemiologic results. Case-control studieson self-reported ultraviolet radiation exposure as a risk factorfor melanoma have been described as especially prone to thedeleterious effect of recall bias because of the growing publicawareness about these risks. Using an ideal test-retest designin a large nested case-control study, Parr et al. (Am J Epidemiol.2009;169(3):257–266) examined to what extent recall biasin melanoma risk factors is actually identifiable and whichconsequences its presence has on effect estimates of these riskfactors. They found only minor indications of recall bias, showingan inconsistent overall pattern and a quite negligible effecton risk estimates. Recall bias was not observed in those exposureswhere it was most expected (solarium use and other ultravioletradiation-related exposures). Their findings cannot be usedas an argument that future case-control studies in melanomaepidemiology should be avoided because of the biasing effectof recall bias. bias (epidemiology); dermatology; epidemiologic methods; melanoma; ultraviolet rays     

Recall bias in exposed subjects following a toxic exposure incident

Attempts to reconstruct health effects following toxic exposure incidents may be subject to bias from distorted or incomplete recollection. We examined recall in 22 of 31 subjects exposed to fumes from ruptured drums containing nitric acid during a U.S. Environmental Protection Agency-supervised hazardous waste site clean up operation in August 1983. Exposed subjects were interviewed by emergency room staff immediately after the episode and the next morning by telephone by a public health epidemiologist. Six months later, the subjects were again interviewed by telephone and asked to report the symptoms they had experienced during the incident. For each respondent, symptoms recalled at 6 mo were compared to symptoms reported at the time. We found a low level of agreement compared to that expected by chance, associated with significant nonrandom differences in the distribution of responses consistently favoring selective recall on the later interview. We conclude that allowing time to elapse before obtaining data on individual symptoms following exposure may lead to a significant bias in response.     

Gender differences in memory for a sexual story

The present research extended previous work that identified gender differences in memory for a sexual text. That work identified a memory bias for the sexes in recognition memory, whereas we found gender differences in errors in recall memory. Recall memory is particularly important because it provides the opportunity for the individual to construct memory. This provides the opportunity for distortion to occur and allows the individual to make errors. The prediction that men would incorrectly recall more material of an erotic nature was supported. The prediction that women would incorrectly recall material that was romantic in nature was not confirmed. In a recognition task both genders endorsed more false positives of a sexual nature than a romantic nature. Using findings from research on memory for written text, predictions concerning the effects of importance, perspective, and typicality were made. Those predictions were not confirmed. A discussion of possible explanations for the various findings is presented.This study is based on portions of a Doctoral Dissertation submitted by J. R. Kirsch to Louisiana State University—Baton Rouge in partial fulfillment of the requirements for the Doctoral Degree.     

Recall bias in self-reported melanoma risk factors

The evidence implicating sun exposure in the etiology of melanoma derives largely from case-control studies in which the retrospective assessment of sun exposure suggests potential for significant recall bias. Previous attempts at characterizing and quantifying that bias have had significant methodological limitations. In the International Twin Study, a case-control study of melanoma risk factors in twins conducted from 1980 to 1991, the authors asked melanoma cases and their co-twins to quantify their own exposures and asked which twin had the greater exposure. Recall bias was investigated by assuming that, if bias had occurred, the odds ratio based on the case's response would differ significantly from the odds ratio based on the co-twin's response. Case-derived odds ratios were higher than the odds ratios for the controls for sunbathing in childhood and adulthood and for mole frequency and freckling in childhood, suggesting some recall bias. The odds ratios for ease of burning and tanning appeared unbiased. The belief that sunlight was a cause of melanoma appeared related to an increased odds ratio for sunbathing as a child. There is a continuing need to carefully assess recall bias in the study of melanoma risk factors.     

Validity of maternal recall for estimating childhood vaccination coverage – Evidence from Nigeria

BackgroundVaccination coverage surveys in low- and middle-income countries typically estimate vaccination coverage using data from vaccination cards, parental recall, or a combination of the two. However, these surveys are often complicated by the pervasive absence of vaccination cards, forcing researchers to rely on parental recall. We assessed the validity of mothers’ recall against home-based vaccination cards using data from a community-based household survey in Nigeria.MethodsA cross-sectional survey of 1,254 mothers of children aged 12–23 months was performed in Enugu State, Nigeria in July 2020. Data on vaccination status for BCG, OPV, DPT, Measles, Yellow fever, and Vitamin A supplement were collected using two data sources: home-based vaccination cards and mothers’ recall. We evaluated the level of agreement between the two data sources; estimated the sensitivity and specificity of mothers’ recalls; and computed multivariable regression models to identify socio-demographic factors associated with mothers’ recall bias.ResultsOut of 1,254 mothers interviewed, 578 (46.1%) mothers with vaccination cards were included in this analysis. Vaccination coverage levels were generally similar across data sources, though recall-based data generally underestimated the coverage. The level of agreement between the two data sources was high (≥91.0% for all vaccine types) with recall bias due to under-reporting generally higher than recall bias due to over-reporting. The sensitivity of parental recalls was high for all vaccine types, while the specificity was low across vaccine types. Across all vaccines, mothers recall bias was significantly associated with the rural residence and not receiving postnatal care.ConclusionIn the absence of vaccination cards, mothers’ recall of their children’ vaccination status for BCG, OPV, DPT, Measles, Yellow fever and Vitamin A is a valid instrument for estimating childhood vaccination coverage in this setting in Nigeria. However, additional research is needed to confirm these findings at higher sub-national and national levels.     

An outbreak of Salmonella enteritidis phage type 34a infection in primary school children: the use of visual aids and food preferences to overcome recall bias in a case control study

Outbreaks of infectious intestinal disease are common in schools. Case control studies are useful in the investigation of infectious disease outbreaks but the time interval between illness and investigation can lead to recall bias, particularly in young children. We describe an outbreak of Salmonella enteritidis phage type 34a infection involving 54 clinical cases in two adjacent schools, and a novel approach to overcome recall bias. The likely dates of infection were identified from the epidemic curve. We created a visual display of the menu from those days and asked 9 cases and 18 matched controls to identify their food preferences from this display. Preference for chocolate mouse was significantly associated with illness (P = 0.006). The results of the case control study agreed with the findings of the environmental investigation. We believe our approach could be used in other circumstances, where subjects are young children or recall bias is a concern.     

Recent life change and large bowel cancer. Data from the Melbourne Colorectal Cancer Study

In a large, population based, epidemiological study of colorectal cancer, The Melbourne Colorectal Cancer Study, several etiological factors were investigated. Persons' recent life changes, as well as the degree of upset they experienced as a result of these changes, were included. Interviews with 715 histologically confirmed new cases of colorectal cancer occurring over a 12-month period in Melbourne, Australia, and with 727 age and sex matched community controls were conducted. As one of the methods of assessing any effect of recall bias, 179 hospital controls were also investigated. Major illness or death of a family member, major family problems and major work problems were found to be significantly more common for cases over the 5 years preceding diagnosis compared to controls. Cases also reported being significantly more upset with their recent life changes than did controls. No significant differences in results were found between males and females, or between colon cancer and rectal cancer patients. Although the possibility of recall bias, was not completely controlled for in this study, it was probably not an important factor in explaining case-control differences. Recent life changes, and their perceptions, may have significance in the development of large bowel cancer.     

The effects of recall on estimating annual nonfatal injury rates for children and adolescents.

OBJECTIVES. This study used a recent national population survey on childhood and adolescent non-fatal injuries to investigate the effects of recall bias on estimating annual injury rates. Strategies to adjust for recall bias are recommended. METHODS. The 1988 Child Health Supplement to the National Health Interview Survey collected 12-month recall information on injuries that occurred to a national sample of 17,110 children aged 0 through 17 years. Using information on timing of interviews and reported injuries, estimated annual injury rates were calculated for 12 accumulative recall periods (from 1 to 12 months). RESULTS. The data show significantly declining rates, from 24.4 per 100 for a 1-month recall period to 14.7 per 100 for a 12-month recall period. The largest declines were found for the 0- through 4-year-old age group and for minor injuries. Rates of injuries that caused a school loss day, a bed day, surgery, or hospitalization showed higher stability throughout recall periods. CONCLUSIONS. Varying recall periods have profound effects on the patterns of childhood injury epidemiology that emerge from the data. Recall periods of between 1 and 3 months are recommended for use in similar survey settings.     

Reporting bias and mode of interview in a study of adverse pregnancy outcomes and water consumption.

In an earlier publication, it was shown that consumption of tapwater during the first trimester of pregnancy was associated with a risk of spontaneous abortion among women who attended three prenatal clinics located in northern California between September 1981 and June 1982. Evidence for biased recall was presented. This report demonstrates that the association is limited to women interviewed by telephone and is not observed among those who completed a mail questionnaire. The possibility that the association among women interviewed by telephone is causal and that the lack of association among mail respondents is due to bias, confounding, or apparently chance fluctuations is shown to be unlikely. For instance, the probability is low that, among mail respondents, cases underreported exposure as compared with controls or that random errors were common enough to camouflage a true effect. Demographic differences between mail and telephone respondents also do not explain the results. Interviewers were, however, aware of the outcome of the pregnancy, suggesting that differential reporting of exposure may have been stimulated in the nonblind telephone interview. The pattern of heterogeneity between mail and telephone respondents was also observed for reportable anomalies and tapwater consumption. The data on heterogeneity support the previously published conclusion that differential reporting (biased recall) of this common exposure, tapwater consumption, is the likely explanation for its observed association with the risk of spontaneous abortion. That this association is limited to telephone respondents elucidates the mechanism that apparently produced the bias: that is, the medium of nonblind telephone interviews.     

A comparison of responses to mailed questionnaires and telephone interviews in a mixed mode health survey

This paper compares respondents to mailed questionnaires with those nonrespondents subsequently interviewed by telephone in a survey of Massachusetts women aged 45-55 years conducted in 1981-1982. This mixed mode approach produced 8,050 responses, giving a response rate of 77%. This rate is similar to rates obtained in many surveys that employed in-person interviews, which are still widely used in health surveys but are increasingly expensive. Telephone respondents differed socioeconomically from mail respondents, suggesting that telephone follow-up of nonrespondents may have reduced nonresponse bias in this survey. Thus, a mixed mode approach may be superior to a mail-only approach with respect to this aspect of data quality. Women responding by mail were more likely to hold professional jobs, to have relatively high household incomes, and to have more years of education. Controlling for these socioeconomic differences did not, however, remove differences in reported health outcomes between mail and telephone respondents. These differences may be explained by less complete recall in the telephone interviews or they may arise from actual differences in health profiles between early (i.e., mail) and late (i.e., telephone) respondents. Although a mixed mode approach may reduce nonresponse bias, more research is required concerning the reasons for response differences between modes and to eliminate any differences caused by problems in data quality.     

Some statistical considerations on dietary assessment methods

Dietary intake for 18 nutrients in a sample of 230 free-living elderly people in Naples was assessed by three dietary assessment methods (diet history, 24-h recall and 3-d record). Different statistical criteria (the correlation coefficient, the mean difference and the proportion of individuals placed in the same thirds of distribution) were used to evaluate the agreement among different methods. Theoretical considerations are shown demonstrating the uselessness of the correlation coefficient in evaluating the agreement. In analysing the mean differences, diet history gave higher estimates of nutrient intake and 24-h recall gave lower estimates, when compared to the 3-d record. Large standard errors for the mean differences and small proportions of individuals placed in the same thirds of the distribution, for nearly all the nutrients examined, showed poor agreement among the three different methods. These findings strongly suggest the presence of bias in, at least, two of the three dietary assessment methods. Some statistical techniques for improving the diet record method in epidemiological studies are discussed: the need to transform nutrient intake data to achieve normality in the distribution; the number of days of recording required to produce a valid study; and the need to evaluate the bias in the diet record.