Association of Anxiety and Unspecified Emotional Distress Obtained from a Medical Records Linkage System with Incident Cognitive Outcomes in a Population-Based Setting

Study design: a prospective cohort study

The MCSA is a population-based study being carried out in Olmsted County, Minnesota, USA. Full details of the study have been published elsewhere.⁹ The study randomly samples individuals residing in this county using an age/sex stratification scheme. While the study focuses on persons that are mid-life and older (i.e., 50 + years in age), sampling does go down to the age of 30 years. If a person agrees to join the study, they participate in a baseline visit and then follow-up visits approximately every 15 months thereafter. The MCSA study approval was obtained from the Institutional Review Boards of the Mayo Clinic and Olmsted Medical Center in Rochester, Minnesota. Participants provided written informed consent before participation. In the case of participants with cognitive impairment sufficient to interfere with capacity, assent was obtained from a legally authorized representative.

In the present study, we included MCSA participants aged 60 years old and older at the time of the current study onset in 2018 (although some were slightly younger than 60 at MCSA baseline), who had undergone at least one neuroimaging study (brain MRI, PiB-PET, FDG-PET), were cognitively unimpaired at MCSA baseline, had neuropsychiatric assessment at baseline (at least one of Beck Anxiety Inventory [BAI], Beck Depression Inventory II [BDI-II], or Neuropsychiatric Inventory Questionnaire [NPI-Q]), and at least one follow-up, and who had not refused permission to use their medical records for research (Minnesota Research Authorization).¹⁰ The present study was approved by the institutional review boards of the Mayo Clinic and the Olmsted Medical Center in Rochester, Minnesota.

Participants’ assessment

All participants went through an exam. During the visit, they were seen by a study coordinator who administered multiple tests including the Beck Anxiety Inventory.¹¹ The study coordinator also administered the Neuropsychiatric Inventory Questionnaire¹² to the participant’s informant. A neurologic exam was performed by a behavioral neurologist. Neuropsychological tests were administered by psychometrists with supervision by neuropsychologists. These neuropsychological tests are grouped into four domains: memory^13,14 (Wechsler Memory Scale – Revised: Logical Memory II Total Score, Wechsler Memory Scale – Revised: Visual Reproduction II Total Score, Auditory Verbal Learning Test: Half Hour Delay), attention/executive^15,16 (Wechsler Adult Intelligence Scale – Revised: Digit Symbol Total Score, Trail Making Test Part B), language^17,18 (Boston Naming Score, Category Fluency Total Score), and visuospatial¹⁵ (Wechsler Adult Intelligence Scale – Revised: Picture Completion Total Score, Wechsler Adult Intelligence Scale – Revised: Block Design Total Score). A z-score for each domain was computed by z-scoring (i.e., subtracting mean and dividing by standard deviation) the individual tests included in a given domain, averaging those z-scores, and then z-scoring that average. A global cognition z-score was computed by z-scoring the average of the four domain z-scores. A consensus diagnosis made by the behavioral neurologist, psychometrist, and study coordinator of cognitively unimpaired (CU),^19 - 22 MCI according to the Mayo Clinic criteria for MCI,^23,24 or dementia²⁵ was assigned. This allowed us to assess the development of incident MCI or dementia. The above participant assessment details can also be found elsewhere with further detail.^26,27 For those that drop out of the study, we were able to follow them passively through the medical record to see if dementia was developed after study discontinuation.

Medical record neuropsychiatric symptoms abstraction

The REP medical record linkage system^10,28,29 is one of the unique resources in the world to investigate the incidence and natural course of several diseases and has been in existence since 1966 with foundations going back to 1907. Using this resource, we were able to abstract data on anxiety and UED. Nurse abstractors trained in extracting data from the REP reviewed and abstracted the data. Details of the REP data resources have been described in detail previously.³⁰

A person was defined as having anxiety the first time an anxiety disorder diagnosis was documented in the patient record; in addition, an anxiety diagnosis was recorded by the abstractor if the patient did not have a documented diagnosis but had anxiety symptoms and received medication (see Supplementary Table 3 for a complete list of medications considered). For the present study, a person was defined as having anxiety the first time anxiety was documented on the record or the first time of medication use for anxiety, whichever was earlier. We also collected information on UED that was not formally diagnosed as a specific psychiatric disorder but was recorded in the medical record in the context of retirement or other life events [i.e., relationship problems, job loss, distress in the context of the daily hassles of life (i.e., conflict at work, social media stress), pet loss, other stress, personality traits (i.e., nervous or anxious personality traits, etc.), or repeated mention over time of stress-related events, etc.]. This allowed us to document emotional distress not reaching the level of the diagnosis of an anxiety disorder. Regarding the timing of UED, we used a similar approach as used with anxiety disorder i.e., a person was defined as having UED at the first time it was found in the medical record, or at the first time medication was used for UED, whichever was earlier. The latter was rarer in the case of UED. There were also cases for both anxiety and UED where exact time of diagnosis was unknown or unclear, and so, time of medication use was used for defining timing of onset. If the first diagnosis came after MCSA baseline, then, for the purposes of this analysis, these neuropsychiatric symptoms were defined as not present. Since we only had year of anxiety diagnosis and year of medication use available, we defined the exact date to be the middle day of the given year (i.e., July 2) for the purposes of comparing to the MCSA baseline date. If anxiety was present but both time of anxiety diagnosis and time of medication use for anxiety were missing, then anxiety was also set to missing for this analysis since we could not determine the timing of diagnosis relative to MCSA baseline. The same was done for UED. Only four had to be set to missing for anxiety and seven for UED. Medical record abstraction was stopped if a diagnosis of dementia was documented in the record.

Statistical methods

Descriptive statistics were calculated and presented as mean (standard deviation [SD]) for continuous variables or frequencies (N) with percentages (%) for categorical variables. In our models, we used two different formulations of the anxiety and UED variables: presence before study entry in the medical record and considering medication use. The variables considering medication use have three levels: No anxiety, anxiety without medication use, anxiety with medication use. The variable considering medication use for UED was defined similarly. As was the case for defining the presence of anxiety and UED variables, the variables considering medication were not able to be computed in cases where medication was indicated but no year of medication use was available. This led to one additional missing value for analyses considering anxiety medication use and two additional missing values for analyses considering UED medication use. We also created a four-level variable for anxiety: No anxiety, anxiety before study entry only, anxiety at study baseline only, and anxiety at study baseline and before. This variable utilized the medical record abstracted anxiety data from before MCSA baseline as well as the BAI and the anxiety item from the NPI-Q collected at MCSA baseline in its calculation. For the BAI, scores 8 and higher were considered as having anxiety (i.e., more than minimal symptoms, indicating clinical anxiety), and presence of anxiety of any severity captured on the NPI-Q was considered anxiety. To be precise, the categories of the four-level variable were defined as follows: 1) No anxiety if anxiety was not found in the medical record before study entry and at least one of the BAI or the NPI-Q anxiety item were filled out at baseline and did not show anxiety; 2) Anxiety before study entry only if anxiety was found in the medical record before study entry and at least one of the BAI or the NPI-Q anxiety item were filled out at baseline and did not show anxiety; 3) Anxiety at study baseline only if no anxiety was found in the medical record prior to joining the study but either the BAI or the NPI-Q anxiety item indicated anxiety at baseline; 4) Anxiety at study baseline and before if anxiety was found in the medical record before study entry and either the BAI or the NPI-Q anxiety item indicated anxiety at baseline. This variable was not able to be computed for one person that was missing both BAI and NPI-Q anxiety data as well as for the four mentioned earlier for which medical record anxiety timing was indeterminate. An analogous variable for UED was not able to be computed as we do not have an in-study measure of UED.

To ascertain the relationship of anxiety and UED with the outcomes of incident MCI and incident dementia, Cox proportional hazards models with age as the time scale adjusting for sex, education, APOE ɛ4 allele status, and Charlson comorbidity index³¹ were run. Linear mixed effects models with participant-specific intercepts and slopes for time were run to estimate the associations of anxiety or UED with longitudinal trajectories of global and domain-specific cognitive z-scores. The models were adjusted for age at baseline, sex, education, APOE ɛ4 allele status, baseline Charlson comorbidity index, and whether this was the first time taking the cognitive test battery. Cox proportional hazards models and linear mixed effects models were also run additionally adjusting for study baseline BDI-II. These results were largely similar, and therefore, not presented, except for one notable difference, which is covered in the discussion. Statistical significance was defined using the usual alpha level of 0.05. Data preparation and statistical analysis was executed using SAS version 9.4 (SAS Institute, Cary, NC) and R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Participant demographics

The sample consisted of 1,808 MCSA participants who were cognitively unimpaired at MCSA baseline. The sample had 930 males (51.4%) and 472 participants were APOE ɛ4 allele carriers (26.1%). The mean (SD) for age at MCSA baseline was 74.5 (7.3) years, and participants had a mean (SD) education of 14.5 (2.8) years and Charlson index of 3.0 (3.0). Anxiety was found in the medical record prior to study entry for 594 (32.9%) participants. This number for UED was 549 (30.5%). Among participants with anxiety, 404 (68.1%) took medication, and among those with UED, 81 (14.8%) took medication. When considering the timing of anxiety, 468 (26.0%) had anxiety before study entry only (as assessed by medical record review), 110 (6.1%) had anxiety at study baseline only (as assessed by BAI/NPI-Q anxiety item), and 125 (6.9%) had anxiety both before study entry and at study baseline (as assessed by medical record review and BAI/NPI-Q anxiety item, respectively). There were 490 (27.1%) individuals who progressed to MCI, and 292 (16.2%) that progressed to dementia. Table 1 contains a summary of the sample.

Results on categorical outcomes of incident mild cognitive impairment and dementia

Participants with anxiety prior to study entry (HR = 1.425, 95% CI: 1.179 to 1.722) had an increased risk of developing MCI at any given time compared to those without anxiety. Presence of UED (HR = 1.160, 95% CI: 0.957 to 1.406) was not found to be associated with increased risk. Anxiety was associated with increased risk of MCI regardless of medication use. The results for the outcome of incident dementia were very similar.

Considering timing of anxiety, participants with anxiety before study entry only, at study baseline only, and at both times all exhibited increased risk of developing MCI. The hazard ratio was the highest for those with anxiety at both times. These results were similar for the outcome of incident dementia with the exception that those with anxiety at study baseline only did not show evidence of increased risk of dementia. Median follow-up time for the outcome of incident MCI was 9 years, and for incident dementia, it was 10 years. All Cox proportional hazards model results can be found in Table 2.

Results on continuous cognitive composite z-scores

The linear mixed effects models revealed that anxiety (Beta=– 0.138, 95% CI: – 0.219 to – 0.058) was associated with lower global cognition z-scores at baseline. Evidence of a baseline association with lower global cognition z-scores was not seen for UED. Participants without anxiety or UED decreased by about 0.09 SD per year in global cognition. Having anxiety or UED accelerated this decline. For example, those with anxiety declined in global cognition by 0.104 (beta for time+beta for interaction=– 0.088– 0.016) SD annually. This increased decline was seen for those not taking and those taking medication for anxiety, although the latter did not reach statistical significance. Increased decline over time was only exhibited for those not taking medication for UED. Anxiety was associated with a faster decline in global cognition for those with anxiety before study entry only and for those with anxiety prior to study and at study baseline, but not for those with anxiety indicated only at study baseline.

The results were largely similar for the outcomes of memory z-score and attention z-score. Anxiety was related to lower z-scores at baseline and faster decline over time, and UED only exhibited longitudinal association. The outcomes language z-score and visuospatial z-score showed some departures from this pattern as anxiety was not associated with faster decline in language or visuospatial z-scores, and UED did not show longitudinal association with visuospatial z-scores. Median follow-up time for the linear mixed effects models was 8 years. All linear mixed effects models results can be found in Table 3 and Supplementary Tables 1 and 2. Figure 1 contains a plot to help visualize the trajectories of the global z-score for the different anxiety timing groups.

Fig. 1

This figure shows a plot of the linear mixed effects model for anxiety timing predicting global z-score. For the purposes of plotting, average values were used for numeric covariates and 0.5 was used for dichotomous covariates.