Stability of individual dementia diagnoses in routine care: Implications for epidemiological studies

Epidemiological and health care research frequently rely on diagnoses from routine care, but the intra‐individual stability of diagnoses of Alzheimer's disease (AD), vascular dementia (VD) or other forms of dementia (oD) in patients over time is understudied. More data on the diagnostic stability is needed to appraise epidemiological findings from such studies.

investigated the intra-individual diagnostic stability of such diagnoses in health claims data over 4 years.
• We found that, depending on the type of dementia, the first and last diagnosis were concordant in between 53.9% and 73.8% of patientswhich means that in up to 46.1% of all patients the diagnosis was changed at least once, with a mean number of changes between 3 and 5.
• Although potentially applicable, dementia-specific interpretations of routine care data warrant caution.

| INTRODUCTION
Dementias rank top among neurodegenerative diseases with an estimated prevalence of 15% beyond the age of 70 years. 1 In addition to the personal burden they also put a substantial burden on society. 2 Patients with dementia have been found to have a markedly increased risk of mortality, with a median survival time of 3-4 years after disease onset as well as an increased risk of institutionalization as compared with age-and sex-matched controls. [3][4][5][6][7] Dementia is associated with substantial costs as well as with an increased caregiver burden, especially in the presence of neuropsychiatric symptoms. 1,8 During the past two decades, the epidemiology of dementia has been extensively investigated in field studies as well as in studies using health claims data. [9][10][11] The latter have been proven to be a valuable data source, since they allow the observation of patients with chronic conditions over a long period of time and across different care sectors. 12 However, most of these studies have assessed dementia as a group of diseases rather than distinguishing between single entities such as Alzheimer's disease (AD), vascular dementia (VD) or other dementias (oD). Consequently, the stability of more specific dementia diagnoses in individual patients over time is understudied. Such data would be helpful to appraise findings from other epidemiological studies, which categorize patients on the basis of a single (either prevalent or incident) diagnosis or two successive diagnoses with a short time interval between them. Thus, potential changes in diagnoses during the course of the disease, which could also have an impact on epidemiological estimators, are therefore usually not taken into account. So far, only one study has addressed this issue by assessing the coding practice of dementia in health claims data over time, revealing that 27% of patients with incident dementia diagnoses received two or more different dementia diagnoses within 1 year. 13 Yet, these important findings were limited by a short observation period and a comparably small random sample of patients who were insured by a regional statutory health insurance (SHI) provider. Moreover, a more in-depth characterization of patients with stable and differing dementia diagnoses is still lacking.
Therefore, based on health claims data from a large sample representing 20% of the German population, the present study sought (1) to estimate the proportion of patients with incident diagnosis codes for AD, VD or oD, (2) to analyze changes in these individual diagnoses over time, and (3) to assess characteristics of patients whose dementia diagnoses as reported in claims records changed as compared to patients with stable diagnoses.

| Study cohort
To be eligible for the study cohort, participants had to be 50 years of age or older, had to be insured anytime between 2004 and 2016, and had to be diagnosed with AD (ICD-10 codes F00: "Dementia in Alzheimer disease" or G30: "Alzheimer disease"), VD (F01: "Vascular dementia") or oD (F02: "Dementia in other diseases classified elsewhere", F03: "Unspecified dementia", or F05.1: "Delirium superimposed on dementia") after at least 1 year without any coded dementia. Moreover, only persons who could be observed for at least 4 years after the incident diagnosis were included.
Participants entered the cohort if they had (a) at least one inpatient dementia diagnosis or (b) at least one outpatient diagnosis coded by a neurologist or (c) one outpatient diagnosis and another in-or outpatient diagnosis of the same dementia category within a time frame of 183 days. Date of cohort entry was the date of incident diagnosis.
Persons with more than one incident dementia diagnosis at cohort entry, e.g., those diagnosed with AD and VD on the same day, were assigned to the dementia category according to the following hierarchy: AD > VD > oD, i.e., patients with incident AD and VD on the same days or incident AD and oD on the same day were assigned to the AD category, patients with incident VD and oD on the same day were assigned to the VD category). Cohort exit was defined by the end of the observation period (i.e., 4 years after cohort entry).

| Definition of diagnostic stability
Upon cohort entry each patient was assigned to the dementia category AD, VD or oD as described above. A change in dementia category was counted if a patient was coded with a dementia diagnosis following assignment to a different dementia category, e.g., code F00 followed by F01 which corresponds to a change from AD to VD. Correspondingly, multiple diagnoses within the same category were considered as stable. Patients with an incident dementia diagnosis only and no further diagnoses in the observation period were not considered for the assessment of diagnostic stability.

| Study measures
Dementia as well as selected comorbidities from the Elixhauser mea-

| Statistical analyses
To assess the stability of dementia diagnoses, the first (at cohort entry) and last available dementia diagnosis during 4 years after cohort entry were considered for each patient. The number of changes between dementia categories was recorded. For each dementia category, absolute numbers and percentages of patients with stable or differing dementia diagnoses were computed. Summary statistics consisted of counts, percentages, means, and quartiles, as appropriate. In order to assess the impact of survival time on the results, all analyses as presented in this manuscript were also conducted with a cohort of patients who could be observed for at least 2 years (see Tables S1 and S2, and Figure S1). Time to first change in dementia category was assessed in Kaplan-Meier curves. All statistical analyses were conducted using SAS statistical software version 9.4.

| Characteristics of the study sample
Overall, 160 273 participants were included in the 4-year observation cohort (see Table 1). A total of 18.8% of them had an incident diagnosis code for AD, 21.5% had an incident diagnosis code for VD, and 59.7% for oD, respectively.
In all three dementia categories, there were more women than men (range: 62.2%-66.7%). Median age was similar in all dementia categories (range: 77-79 years). Among patients with incident diagnosis codes for AD, 9.8% had at least a diagnosis from another dementia category on the same day and thus were assigned based on the hierarchical rule (for VD: 3.5%). Irrespective of dementia category, the incident diagnosis code was either coded by a neurologist or a GP in more than half of all patients, and the most common comorbidity was essential (primary) hypertension affecting between 61.6% and 69.7% of all patients, followed by disorders of lipoprotein metabolism and other lipidemias (39.1%-42.2%). Furthermore, between 24% and 25.2% of patients had diagnosis codes for depression. Overall, patients with incident diagnosis codes for VD suffered from diseases of the circulatory system at baseline more often than those with incident diagnosis codes for AD or oD, respectively. The cohort of the 2-year observation cohort comprised of 310 938 participants and had similar characteristics like participants in the 4-year observation cohort (see Tables S1 and S2 for further details).

| Diagnostic stability
Of 160 273 patients in the 4-year observation cohort for 15 842 was no further dementia diagnosis coded after their incident diagnosis.
The proportion of patients with only one coded diagnosis was higher in patients with incident diagnosis codes for VD (12.4%) and lower in patients with incident diagnosis codes for oD (9.8%) or AD (7.4%), respectively. Thus, 144 431 patients with more than one dementia diagnosis were included in the analyses of diagnostic stability.   The same proportion was reached after 506 days in the VD subgroup and after 733 days in patients with incident diagnosis codes for oD.

| Proportions of dementia diagnoses
One in five patients was diagnosed first with AD or VD while more than half of all patients were initially diagnosed with oD. These proportions are substantially lower than the 60%-70% of AD cases among all dementia cases repeatedly reported in the literature. 16 In fact, findings from the Rotterdam study which were derived from a small community-based sample in the early 1990s, 17  main medical contact for patients after diagnosis. While this is not generally ruled out by established guidelines, 20 it is also not specifically recommended, especially in the presence of neuropsychiatric comorbidities, which are common in dementia. 21 Again this could also hint at some patients already living in nursing homes where routine examinations are usually conducted by GPs rather than specialists. It is interesting though to note that the specialist treatment rates are similarly low for patients with and without changes in diagnosis. This suggests that consulting a specialist does not lead more often to alterations (also in terms of corrections) of incident diagnoses.

| Implications and limitations
In summary, the results of our analyses show that in well over half of all dementia patients, the incident diagnosis remains stable and is confirmed during the course of the disease, possibly multiple times.

ETHICS STATEMENT
In Germany, the utilization of health insurance data for scientific research is regulated by the Code of Social Law. All involved health insurance providers as well as the German Federal Office for Social Security and the Senator for Health, Women and Consumer Protection in Bremen as their responsible authorities approved the use of GePaRD data for this study. Patient consent was not required for this study.

PRIOR POSTINGS AND PRESENTATIONS
Not applicable.