Cost-effectiveness of comprehensive geriatric assessment at an ambulatory geriatric unit based on the AGe-FIT trial

The aim of this study is to assess the cost-effectiveness of CGA at an ambulatory geriatric unit, compared to usual care for elderly people with multi-morbidity and high healthcare consumption.

The patient population evaluated comprised elderly people with multi-morbidity and high healthcare consumption according to the AGe-FIT trial [8, 9]. The evaluated treatment strategies were CGA at an ambulatory geriatric unit as in the AGe-FIT trial described below and usual care. Costs and QALYs, weighting each year lived with the quality of life (where 1 is full health and 0 is dead), were evaluated over a lifelong time horizon for each treatment strategy using standard methods of decision-analytic modelling [11], synthesizing two years of trial data from the AGe-FIT trial with other relevant data sources. The primary outcome was the incremental cost-effectiveness ratio (ICER) of CGA compared with usual care, relating the incremental costs of CGA to the incremental health outcome (QALYs):

In addition, cost per life year (without quality adjustment) gain was assessed. Costs and health outcomes were discounted by 3% per annum. All costs were calculated in Swedish kronor (SEK), and converted to 2016 euros (€) using the exchange rate of 1 EUR = 9.47 SEK (year 2016 mean exchange rate). The analysis was performed from the perspective of the public healthcare sector including costs for services in both the Municipality and the County Council.

The main data source for the present study, the AGe-FIT trial, has been reported in detail elsewhere [8, 9, 12]. In brief, the AGe-FIT trial was randomized, controlled, assessor-blinded, and carried out in the Municipality of Norrköping, Sweden during 2011 and 2013. Community dwelling patients, of 75 years or older, having three or more concomitant diagnoses and admitted to the hospital for inpatient care three times or more during the past 12 months, were included. The participants gave written informed consent to participate in the study, in case of cognitive decline a proxy gave informed consent according to the protocol [12]. Participants who gave written consent to participate were randomized to either the intervention group receiving care at the ambulatory geriatric unit (AGU) in addition to usual care, or the control group receiving usual care only. The intervention consisted of CGA provided at the AGU and included an interdisciplinary approach in order to make a person-centered plan for future care [12]. Most often it started with a home-visit from a nurse and a social worker to focus on “caring-problems” such as nutrition, skin-problems, elimination and the social environment. A pharmacist reviewed medication before a visit to a physician in the AGU and, depending on needs, a functional investigation by the occupational and physical therapists. In the usual care strategy, standard healthcare services were used as needed.

The AGe-FIT trial was approved by the regional ethical vetting board at Linköping University (No: 2011/41–31 and No: 2015/6–32), and is registered on http://​clinicaltrails.​gov (NCT01446757). The study adheres to the CONSORT guidelines [13].

In order to estimate the long-term costs and QALYs and assess cost-effectiveness of the CGA intervention, a simple two-state (Alive and Dead) Markov model was employed [14]. In the model, all patients start in the Alive state. During each year (annual Markov cycle) patients face a risk of dying, and thus transition to the absorbing Dead state. The annual risk of dying was conditional on the assigned treatment strategy and based on the 24-month follow-up data from the AGe-FIT trial. The model was run for 30 cycles to ensure that in effect all patients had reached the absorbing Dead state at the termination of analysis. For each Markov cycle that patients reside in the Alive state, they incurred an annual cost and QALY-estimate. There were no costs and QALYs associated with the Dead state. At the termination of analysis, discounted costs and QALYs were summed over all cycles to estimate per patient mean costs and QALYs for CGA and usual care, respectively.

An overview of data inputs is provided below. Further information is provided in Additional files 1 and 2.

Data on healthcare resource use was collected for primary healthcare, ambulatory care (geriatric and other), inpatient care and municipal services (including use of home help services and nursing home utilization) in the Age-FIT trial by linking trial participant–ID to relevant patient registers [8]. For the first two years (Markov cycles) of the analysis these costs corresponded to the per patient mean cost per treatment strategy observed in the AGe-FIT trial (Table 1). From year three and onwards the costs observed in year two of the AGe-FIT trial were applied to the Alive state. The impact on the final results of extrapolating the observed two-year data over a longer time horizon was explored in sensitivity scenarios. See Additional file 2 for details.

A QALY-estimate for the Alive state, per treatment strategy, was based on HRQoL data measured with the EQ-5D-3L instrument in the AGe-FIT trial at baseline, 12 months and 24 months [15]. Patient answers on the EQ-5D-3L instrument were converted to QALY-weights using the widely used UK value set [15]. QALY-estimates for year one and two were estimated by calculating the area under the curve for patients alive at 12 and 24 months (Table 1). The QALY-estimate for year two was applied to the Alive state for year three onwards. Furthermore, an annual reduction (for both treatment strategies) in HRQoL was applied to account for the fact that patients get older with an expected health deterioration. Based on published EQ-5D data on the Swedish general population, the QALY-estimates were adjusted by a decrement of 0.0025 per year [16]. The impact on the final results of adjusting the long-term QALY-estimates, was explored in sensitivity scenarios.

For the first two years of the analysis the mortality risks associated with each treatment were estimated from the AGe-FIT trial, and thus corresponded to the observed mortality in the trial. For the first year of analysis this risk was 11.5% and 13.8% for CGA and usual care, respectively. Corresponding figures for year two were 8.2% and 15.3%. The mortality risks for year three and onwards were estimated by using age-specific mortality rates for the general population in Sweden [17]. See Additional file 1 for details.

The analyzed patient population corresponded to the characteristics of patients in the Age-FIT trial with the mean age of the population in the trial (83 years). Uncertainty in the estimated cost-effectiveness results associated with sampling uncertainty in the estimated input parameter values, was evaluated by employing probabilistic sensitivity analysis [18]. In this analysis, the uncertainty in single-model inputs is propagated through the model, using simulation techniques, so that the uncertainty in the cost-effectiveness results indicates the uncertainty in the decision to implement a treatment strategy, rather than the uncertainty surrounding single model inputs [18]. The probability of the CGA being cost-effective at different threshold values for cost-effectiveness was assessed and reported in cost-effectiveness acceptability curves [19]. The importance of parameters not associated with statistical uncertainty for the final results were investigated in sensitivity analyses. See Additional file 2 for details.

All statistical analyses of Age-FIT trial data were performed in SPSS version 22.0 [20]. The decision-analytic model was programmed and analyzed in Microsoft Excel (Microsoft Corporation, Redmond, Washington DC, USA).