Table 2.

Attributable Cost of Dementia among Veterans Age 65 and Older Who Used Veteran Health Administration (VHA) Using VHA Data Alone or Combined VHA and Medicare Data in FY 2013

	VHA Data Alone†		Combined VHA and Medicare Data‡
	Diagnosis in VHA(N = 92,970)		Diagnosis in VHA or in Medicare(N = 142,951)
	Unadjusted Attributable Cost	Adjusted Attributable Cost	Unadjusted Attributable Cost	Adjusted Attributable Cost
To VHA	$15,222*** (15,107–15,336)	$10,950*** (10,916–10,984)	$9,945*** (9,850–10,039)	$6,662*** (6,650–6,674)
To Medicare			$14,003*** (13,896–14,111)	$5,522*** (5,509–5,535)
To combined VHA and Medicare	–	–	$24,558*** (24,428–24,688)	$11,285*** (11,263–11,306)

The study cohort is veterans age 65 and older who used VHA and were enrolled in Medicare but not Medicare Advantage in FY 2013.

We use GEE models with gamma distribution and natural logarithm link and exchangeable working correlation with variance clustered at VAMC. The models are offset by the natural logarithm of the number of days alive in FY 2013.

Recycled predictions are used to calculate attributable cost of dementia based on estimates from GEE models: (1) predict health care costs for each veteran assuming the veteran had dementia and assuming the veteran did not have dementia, holding all other characteristics as observed; (2) calculate the incremental cost attributed to dementia for that veteran by taking the difference between the predicted costs assuming the veteran had dementia compared to the predicted costs assuming the veteran did not have dementia. Attributable cost of dementia is calculated as the sample average of all veterans’ incremental cost attributed to dementia. Adjusted attributable cost of dementia and 95% CI are reported.

Unadjusted attributable costs and 95% CI were reported. T‐test was used to compare the differences in unadjusted costs between patients with dementia and patients without dementia.

*Significant at 10%; **significant at 5%; ***significant at 1%.

VHA and Medicare costs were CPI adjusted to 2016 with Wage Index (2/3) and truncated at the 99th percentile.

Control variables include demographic variables (age groups, gender, marital status, race), rurality of residence (highly urban, urban, rural, highly rural), JFI scores, Medicaid eligibility and market characteristics (% population ≥75, # nursing home beds per 1,000 population, # hospital beds per 1,000 population and # active physicians per 1,000 population in the county, and % state Medicaid spending on HCBS).

†Under “VHA Data Alone” column dementia, risk scores and comorbidities were computed using VHA data alone.

‡Under “Combined VHA and Medicare Data” columns dementia, risk scores and comorbidities were computed using combined VHA and Medicare data.