Time Trends in the Prevalence of Neurocognitive Disorders and Cognitive Impairment in the United States: The Effects of Disease Severity and Improved Ascertainment

Abstract

Background: Trends in the prevalence of cognitive impairment (CI) based on cognitive assessment instruments are often inconsistent with those of neurocognitive disorders (ND) based on Medicare claims records.

Objective: We hypothesized that improved ascertainment and resulting decrease in disease severity at the time of diagnosis are responsible for this phenomenon.

Methods: Using Medicare data linked to the Health and Retirement Study (1992-2012), we performed a joint analysis of trends in CI and ND to test our hypothesis.

Results: We identified two major contributors to the divergent directions in CI and ND trends: reductions in disease severity explained more than 60% of the differences between CI and ND prevalence over the study period; the remaining 40% was explained by a decrease in the fraction of undiagnosed individuals.

Discussion: Improvements in the diagnoses of ND diseases were a major contributor to reported trends in ND and CI. Recent forecasts of CI and ND trends in the U.S. may be overly pessimistic.

Keywords: Alzheimer’s disease; Ascertainment; Health and Retirement Study (HRS); Medicare; Telephone Interview for Cognitive Status (TICS); cognitive impairment; disease prevalence; neurocognitive disorders; severity; time trends; underdiagnosis.

Figure 1.

Annual percent change (APC) of age-adjusted indicators I(T ≤ T_thr) of low TICS scores using values for the TICS threshold cut-point (T_htr) in the range 2–27.

Figure 2.

Time trends of age-adjusted prevalence for selected diseases from the memory disease group. Each point represents ND prevalence at the mean date of the interview of a specific wave of HRS data.

Figure 3.

Mean values and error bars for standard errors of TICS scores for individuals with any of 20 diseases. Closed and open dots correspond to original measurements (i.e., without proxy respondents) and filled (with imputed data) data, respectively. The first two columns in the right side of this figure show the number of measurements of TICS for individuals with a given disease for the filled (column 1) and base (column 1) datasets. The differences between them are the number of proxy respondents. The next three columns show the fractions of proxy respondents with 1+, 2+, and 3+ negative responses.

Figure 4.

Age-adjusted prevalence of CI (open dots) and ND (closed dots) (p_CI and p_ND) and the relative frequencies of the two-way analysis f₁₁ (thin solid), f₁₀ (dashed), and f₀₁ (thick solid). The second header line in each panel shows the type of CI imputation (Base or Filled), the threshold used, and the percent explained difference between CI and ND trends calculated, respectively, without and with additional adjustments for sex and education. NA for senile dementia means that the difference between CI and ND prevalence was not significant.