Assessing the Reliability of Reported Medical History in Older Adults

INTRODUCTION

A reliable medical history is critical to prioritize investigations, establish accurate clinical diagnoses, and select appropriate therapies [1–3]. Conversely, inaccuracies in the medical history contribute to diagnostic errors, inflate the cost of medical care, and increase the risk of morbidity, mortality. and medical malpractice claims [4, 5]. Although many sources may inform the medical history, direct patient and caregiver reports remain the primary source of information. Patient and caregiver reported history is especially important when considering conditions defined by interval changes in health status (e.g., dementia [6]), or when determining eligibility and allocation within observational studies and clinical trials. Despite its importance to clinical care and research, few studies have rigorously considered the factors that influence the reliability of direct reporting. Those that have vary in the use of reference standards, yielding conflicting results [7–10].

Inaccuracies in reporting of medical history may arise from many sources, including errors in reporting due to misunderstandings or misinterpretation of past results, and lack of knowledge concerning an underlying risk factor or health problem. Older adults may be particularly susceptible to inaccurate reporting and associated sequelae owing to age-associated increases in medical complexity, frailty, and cognitive impairment [11, 12]. It is particularly important to identify the factors that influence the reliability of individual-reported medical history in this vulnerable population. Recognizing this, we evaluated the concordance between reported history of stroke and diabetes, and detection of cerebral infarction on magnetic resonance (MR) neuroimaging and elevated hemoglobinA1c (HbA1c) in community-dwelling older adults enrolled in longitudinal studies of memory and aging at the Knight Alzheimer Disease Research Center (ADRC; St. Louis, MO, USA). Stroke/cerebral infarction and diabetes were selected as exemplar variables due to their prevalence in the general population, presumed relationship with Alzheimer’s disease pathophysiology and dementia risk [13], potential to present or develop insidiously, and the availability of sensitive and specific measures of both disorders (MR neuroimaging for detection of cerebral infarcts [14]; HbA1c for detection of diabetes mellitus [15]), permitting comparison of reported history with objective reference standards in research participants. We hypothesized that the reliability of reporting of medical history would be influenced by key participant-specific variables, including age and cognitive impairment.

METHODS

RESULTS

Data were available from 1,401 unique participants assessed from April 2005 to April 2019. Participants were well-educated (mean 15.7±2.7 years of education) older adults (mean 73.4±9.2 years-of-age). Racial distribution was consistent with the surrounding community: 80.5% (n = 1128) of participants were non-Hispanic white; 18.3% were African American (n = 256). The majority of participants were cognitively normal (CDR = 0, n = 976 : 69.7%). When present, cognitive impairment was generally very mild (CDR 0.5, n = 262 : 18.7%) or mild (CDR 1, n = 138 : 9.9%), with symptoms most commonly attributed to Alzheimer’s disease (324/425, 76.2%). Magnetic resonance neuroimaging was performed in 862 participants and HbA1c measured in 1,319 participants. Demographic and clinical characteristics are reported in Table 1 for each cohort.

The concordance between reported history of stroke and detection of past ischemic or hemorrhagic infarction on neuroimaging was low, with a sensitivity of 17.4% [95% _CI 6.5–28.3%] (8/46; Supplementary Table 1). A prior history of stroke was reported by 25/862 (2.9%) participants, yet confirmed on neuroimaging in only 8 of these 25 participants (PPV = 32% [95% _CI 16.0–52.0%]). The specificity of participant report was 97.9% (799/816 [95% _CI 96.9–98.9%]). Infarcts were detected on neuroimaging in 38 participants who did not report a past history of stroke (negative predictive value = 95.5% [95% _CI 94.0–96.8%], 799/837). 11/46 infarcts (26%) were hemorrhagic; all were small (≤9 mm³). Microhemorrhages were detected on susceptibility-weighted MR neuroimaging in 97 participants (11.2%). Older age, the presence of cognitive impairment (CDR≥0.5), greater degree of cognitive impairment (lower MMSE scores), history of hypertension and use of an unrelated collateral source (i.e., close friend or acquaintance) were more frequent in participants with discordant reporting (Table 2).

When these variables were considered together via multivariable logistic regression, the presence of small infarcts emerged as the predominant driver of inaccurate reporting (odds ratio [OR] = 265.8 [95% _CI 86.2, 819.4]; controlling for age and sex). A potential clinically meaningful association between an unrelated collateral source and discordant reporting could not be excluded (OR = 2.4 [95% _CI 0.99–5.9]; n = 860; Supplementary Table 2). Cognitive impairment (OR = 1.8 [95% _CI 0.63, 5.0]) and increasing age (OR per decade = 1.2 [95% _CI 0.70, 2.1]) were not associated with increased odds of discordant reporting. The multivariable model demonstrated a good ability to identify participants in whom reported stroke history and neuroimaging findings were most likely to be discordant (c-statistic = 0.87 [95% _CI: 0.81, 0.93]; p < 0.001).

The sensitivity and specificity of reported history of diabetes was 83.5% [95% _CI 77.8–88.6%] (147/176) and 96.2% [95% _CI 95.1–97.3%] (1100/1143), respectively (Supplementary Table 3). A history of diabetes was reported by 190/1319 (14.4%) participants. Of these, 147 participants had an HbA1c consistent with diabetes (≥6.5) or were taking medications for the management of diabetes (PPV = 77.4% [95% _CI 71.1–83.2%]). Serum HbA1c was consistent with diabetes in an additional 29 participants who denied a past history of diabetes and who were not taking medications commonly prescribed to treat diabetes (NPV = 97.4% [95% _CI 96.5–98.3%]). Older age, African American race, a history of hypertension, higher HgA1c, and HgA1c in the range suggesting impaired glucose tolerance (6.0–6.4) were associated with discordant reporting of diabetes (Table 3).

When these variables were considered together via multivariable logistic regression, a history of hypertension (OR = 2.3 [95% _CI 1.3, 4.2]), higher HgA1c (OR = 1.9 [95% _CI 1.5, 2.4]) and HgA1c compatible with impaired glucose tolerance (OR = 3.1 [95% _CI 1.8, 5.3]) were independently associated with increased odds of discordant reporting of diabetes (n = 1318; Supplementary Table 4), when controlling for sex. Cognitive impairment (OR = 1.0 [95% _CI 0.6, 1.7]) and increasing age (OR per decade = 1.1 [95% _CI 0.80, 1.5]) were not associated with increased odds of discordant reporting. The multivariable model demonstrated fair ability to identify participants in whom reported history of diabetes and HbA1c measures were most likely to be discordant (c-statistic = 0.79 [95% _CI 0.74, 0.84], p < 0.001).

False negative reporting accounted for the majority of discordance between stroke history and neuroimaging findings (69%, 38/55), indicating that occult (i.e., clinically silent) infarction accounted for the majority of discordant reports. The opposite was true concerning diabetes: participants were more likely to over-report a history of diabetes (“false positive”), despite an HbA1c in the normal range (<6.5%). The factors associated with false positive and false negative reports were explored through post-hoc analyses (Table4). Female participants were more likely to over-report (14/17, 82%) than under-report (20/38, 53%; p = 0.04) a history of stroke. No other differences were noted in participant-specific factors, cognitive function, medical history, or other measured variables.

Both neuroimaging and HbA1c data were available in 780 participants. No relationship was observed between discordant reporting concerning history of stroke and diabetes (inter-reporter agreement, Cohen’s kappa = 0.03; p = 0.46); i.e., discordant reporting in one domain did not predict discordance in the other.

DISCUSSION

Discordance between participant reported history of stroke and diabetes and objective measures of cerebral infarcts and chronic elevations in blood glucose were common in community-dwelling older individuals enrolled in longitudinal studies of memory and aging at our center. More than two-thirds of participants who endorsed a prior history of stroke did not have cerebral infarcts on neuroimaging, while >80% of participants with imaging-confirmed infarcts did not endorse a history of stroke. The presence of small infarcts on neuroimaging was strongly associated with discordant reporting, with abundant “false negative” reports, implying that the majority of cerebral infarcts were undiagnosed, unrecognized or occult/clinically silent. Participants were more reliable when reporting a history of diabetes. Nonetheless, HbA1c was not consistent with diabetes in almost one in four participants who endorsed a history of diabetes, and 16.5% of participants were unaware that they had diabetes. HbA1c levels consistent with impaired glucose tolerance (6.0–6.4%) were independently associated with inaccurate reporting, suggesting that participants may have misinterpreted their diabetic status, contributing to false positive reports. A history of hypertension was also associated with greater odds of inaccurate reporting of diabetes history. This may reflect heightened awareness or concern regarding diabetes in participants with other “vascular risk factors” who may be more likely to receive routine blood glucose screening, and counseling concerning the need to manage impaired glucose tolerance and prevent diabetes [21]. This possibility emphasizes the potential for confusion and misinterpretation of information provided by care providers, and the need for increased counseling concerning the implications of “high risk” or prodromal disease states (e.g., impaired glucose tolerance, which may be presented as “early” or “pre-diabetes”) and the differences between these transitional states and diagnoses associated with adverse outcomes (e.g., diabetes).

Cognitive impairment and increasing years of age were not associated with reporting accuracy. These findings differ from reports in elderly persons living in northern Manhattan, in whom older age and impaired memory/cognitive skills associated with greater odds of inaccurate self-reporting of stroke [8]. These differences may reflect the generally low-level of cognitive impairment in our participants, recognizing that the majority of participants were very mild or mildly affected (CDR 0.5/1.0). Additionally, the incorporation of a reliable collateral source in the medical interview may have blunted the influence of cognitive impairment on reporting accuracy in our study. In support of this suggestion, a clinically meaningful association between the participant-collateral source relationship and odds of discordant reporting could not be excluded (OR = 2.4 [95% _CI: 0.9–5.9]). Inclusion of a reliable collateral source in the medical interview may facilitate engagement and comprehension of shared information [22, 23], benefits which may be most apparent when assessing cognitive function in individuals with early-stage dementia [24–26] or when evaluating sicker individuals [23].

Reporting accuracy was higher for diabetes than “stroke”, with fewer false negative reports. This difference may reflect routine screening for diabetes in individuals > 45 years of age (consistent with recommendations from the American Diabetes Association [15]), facilitating the detection of diabetes in asymptomatic individuals. The chronic nature of diabetes may also have contributed to accuracy, noting that the concordance between self-report and objective measures of disease are generally higher in well-defined conditions that require frequent monitoring (e.g., daily serum glucose checks) and regular use of medications (oral or injection) [27]. Conversely, stroke and cerebral infarction are the exemplar acute disease states that may present with symptoms and signs that are easily overlooked or misinterpreted by patients and providers [12, 28] or with no symptoms at all (silent infarction). Asymptomatic presentations may be even more common when infarct volume is small, contributing to false negative reporting in this study and others [29, 30], and emphasizing the potential for cerebrovascular disease to insidiously progress in older individuals. Selected factors under conscious and unconscious control that may contribute to errors in reporting of medical history are summarized in Table 5, along with strategies that may be leveraged to mitigate discordant reporting of medical history in clinical and research settings.

Beyond informing the factors that influence concordance between historical reports and objective measures of disease, the findings from this study have direct implications for clinical care and research. As outlined in North American screening recommendations [21], the early detection of diabetes may provide an opportunity to intervene before symptoms or debilitating multi-organ consequences of diabetes manifest. In the same way, the detection of occult infarcts, including small incidental infarcts, should trigger the clinician to screen for and treat other vascular risk factors, with the goal of preventing future strokes and improving outcomes [34]. Accurate recognition of cerebral infarcts and diabetes may be particularly important when assessing individuals with cognitive decline, recognizing the relationship between cerebral infarcts and vascular cognitive impairment [35, 36], and the potential contributions of vascular risk factors and cerebrovascular disease to the age-at-symptomatic onset, phenotype and rates of progression of common age-related neurodegenerative diseases, including Alzheimer’s disease [13, 37–39]. In this context, our findings highlight the need for caution before acting on historical reports. Stated simply, the presence or absence of cerebral infarcts or diabetes should be objectively confirmed when the findings will be used to direct clinical care, inform diagnostic classification or influence study inclusion/exclusion.

This nested cohort study leveraged data from a robust number of participants enrolled in studies of memory and aging at a single center, with racial breakdown representative of the local population. Prospective collection of information from diverse cohorts and the ability to reference “gold standard” measures for cerebral infarction and persistently elevated blood glucose are notable strengths. However, our results are subject to limitations. Analyses of reporting reliability were limited to diabetes and stroke. Thus, it is unclear whether our findings can be generalized to other diseases (e.g., myocardial infarction) or aspects of the medical history (e.g., medication use). In particular, accurate reporting of diagnoses or habits associated with perceived stigma (e.g., mental illness, substance use) may be influenced by other factors that were not measured in this cohort, warranting further study. Our participant population was relatively well-educated with a lower prevalence of diabetes and symptomatic stroke than is reported in the United States and other community-based studies [8, 41]. Additionally, enrolled participants were willing to participate in annual clinical and neuropsychological assessments, with frequent biofluid sampling and neuroimaging. Higher levels of engagement in research are likely to associate with higher degrees of health literacy and improved access to care. Reporting accuracy may be lower in less-engaged/motivated individuals, and in those from disadvantaged populations, including communities with less access to recommended screening, lower health literacy, and more prevalent medical comorbidities. As a result, our findings likely underestimated the potential for discordance between participant-reported history of stroke and diabetes and objective measures in the general population.