Cost-Effectiveness of Comprehensive Geriatric Assessment Adapted to Primary Care

hospitalization pragmatic

Many countries in the world are faced with the major challenge of managing increased health care needs among older adults.This is partly caused by an aging population and increasing incidence of frailty and multimorbidity. 1,2Moreover, expectation of a healthy and active life and new treatments further increase the gap between what is possible and the resources available for health care. 3In this context, prioritization is necessary in all parts of the health care system.The ability to succeed and make wise decisions regarding priorities is dependent on more evidence concerning the cost-effectiveness of treatments and care models.
There is an ongoing debate in research about how frailty should be defined, and various models have been proposed. 4The World Health Organization report on healthy aging describes frailty as decreased intrinsic capacity (the composite function of the various organ systems) that makes the individual vulnerable to various stressors, with a risk of rapid loss of function. 3Frailty is related to aging and multimorbidity but is not necessarily a consequence of these.Frailty is also described as a dynamic condition that can be delayed and, to a certain extent, reversed. 4Economic evaluations have demonstrated health care costs for frail older adults that are 2 to 3 times higher than those for robust individuals. 5The predominant sources of health care costs for frail individuals are hospitalization and post-acute care. 6n the absence of firm evidence of effective treatments and care models for managing frailty, there are several recommendations that highlight a holistic care strategy for frail older adults. 3,7,8This includes a recommendation for comprehensive geriatric assessment (CGA) and the formulation of individualized and proactive care plans that encompass the values and priorities of the older adult. 9Interventions using CGA in hospital and post-acute settings have demonstrated significant positive effects on survival and decreased admission to nursing homes, but there is still insufficient evidence. 10A few studies also have demonstrated the cost-effectiveness of these interventions. 11,12rimary care has a role as the initial contact point and by providing continuity of care over time for all general health issues in the population, and recommendations support the notion that primary care should be the first-line management option for people with frailty and multimorbidity using comprehensive care models like CGA. 8 A lot of research has been conducted over the past 20 years and several comprehensive models for primary care have been evaluated 13e15 ; however, there is still no convincing evidence for effective comprehensive care strategies, despite these numerous studies, and only a small number of them have evaluated the cost-effectiveness of the interventions.16e20 These evaluations have mostly demonstrated higher costs in the intervention group and either small gains in functional ability or no significant effect compared with care as usual.
Comparisons between studies are difficult, as the interventions have used a variety of outcome measures and different follow-up periods.
In Sweden, the research project "Proactive Healthcare for Frail Elderly Persons" studied the effects of CGA in primary care in a group of vulnerable older adults who were identified using a prediction model. 21redicting risk of hospital admission has been suggested as a way to identify vulnerable older adults without having to manually assess the person at a clinical appointment.22e24 The intervention in "Proactive Healthcare for Frail Elderly Persons" demonstrated a relative risk reduction of 22% for hospital care days and 17% lower total health care cost compared with care as usual. 25Therefore, we considered it important to also evaluate the cost-effectiveness of the intervention.
The aim of this study was to analyze the cost-effectiveness of a CGA intervention adapted to primary care delivered to a group of older adults at high risk of hospitalization.

Design
The present study is a within-trial cost-effectiveness analysis.The original study "Proactive Healthcare for Frail Elderly Persons" was a pragmatic matched-controlled trial at 19 primary care practices in southeast Sweden with follow-up over 24 months that has been described elsewhere. 25,26

Participants
In the original study, we selected 1604 participants aged 75 years and older using a prediction model that calculates a risk score for hospitalization in the next 12 months using routine health care data.A total of 1308 participants were alive at the start of the follow-up period.A questionnaire was sent by mail to all participants on 3 occasions during the study, at baseline, at 10 months of follow-up, and at 22 months of follow-up.In connection with the baseline questionnaire, participants were asked for their consent to analyze their answers together with their health care utilization.In total, 369 individuals agreed to participate and were included in the present analysis.The study was registered at ClinicalTrials.gov(Identifier: ctgov:NCT03180606, first posted August 6, 2017) and was approved by the Regional Ethical Review Board in Linköping (Reg.no.2016/347e31).

Intervention
Participants at the 9 practices involved in the intervention were invited to undergo CGA performed by primary care nurses.The assessment was performed using a new CGA tool; Primary care ASsessment Tool for ELders (PASTEL). 27After the assessment, the nurse met with the responsible physician to jointly estimate the participant's degree of frailty and to plan further investigations and actions.All follow-up actions and activities were individually tailored; there was no standard treatment or follow-up.The assessment and care planning took place during the run-in period that lasted 9 months (April to December 2017) and the subsequent follow-up period lasted 24 months (January 2018 to December 2019) The 10 control practices were matched to the intervention practices with respect to the number of registered older adults and sociogeographic location.The control practices provided care as usual.

Outcomes Health-related quality of life and mortality
Health-related quality of life (HRQoL) was measured using the EQ-5D-3L instrument including EQ-5D-VAS, and was obtained from the questionnaires sent at baseline, and at follow-up after 10 months and 22 months. 28We used the UK value set to convert the participants' answers to the EQ-5D index representing their health state. 29These scores range from À0.594, representing lowest quality of life, to 1.00, representing full health.
Date of death was obtained from the Swedish Tax Agency's population register.
We calculated quality-adjusted life years (QALYs) by multiplying the time spent in a particular health state with the corresponding EQ-5D index (QALY weight) and then added the 3 periods to a sum of QALYs for the entire follow-up period.We considered the index value to be stable until the next measurement point.

Costs
The health care costs in the follow-up period were calculated using the care data warehouse linked to the cost-per-patient database of Region Östergötland.The care data warehouse contains all health care contacts for both public and private care providers and the cost-perpatient database includes total costs for all contacts within public health care.The cost calculations used in this study have previously been reported in more detail. 25The costs of the intervention incurred during the run-in period (ie, introduction and education of the health care staff, together with time spent on assessments and team meetings) were estimated by the research group as the total hours spent by nurses and physicians at a primary care practice of average size.We then divided the sum total by the total number of participants at that practice.Gross salaries for physicians and nurses were obtained from the region's register.
The cost of home help services and nursing home costs were obtained from the questionnaire.Participants were asked to report the number of hours of home help services per week at baseline and the 2 follow-ups.They also reported the type of housing they were living in.We considered the reported hours of home help services and type of housing to remain unchanged until the next follow-up or death.We used an average price per hour for home help services and per day for nursing homes in the municipalities across the region, as reported to the Swedish Association of Local Authorities and Regions.All costs were converted to US dollars (USD 1 ¼ SEK 10)

Cost-effectiveness analysis
We analyzed the cost-effectiveness using 2 perspectives: the health care perspective, including costs for primary and secondary health care, and the societal perspective, which also included community costs for home help services and nursing home care.Incremental cost-effectiveness ratios were calculated for both the intervention group and care-as-usual groups by calculating cost/ QALYs gained during the 24-month follow-up period.Adjusted data  concerning costs and effects were used in the cost-effectiveness analysis, and the data were bootstrapped through 10,000 iterations.Adjustments were made for age, gender, and risk score.The uncertainty of the cost-effectiveness analysis is described in 2 different costeffectiveness planes that illustrate the 2 perspectives.

Statistical analysis
Because of missing questionnaires and missing data in included variables, a complete case analysis would have excluded at least 30% of the initial cohort, potentially introducing a bias if the excluded cases were a nonrandom sample.We therefore used the multiple imputation by chained equations (MICE) package in R to deal with missing data for those patients still alive at different time points.We used n ¼ 10 imputed data sets.In the imputation modeling we included data concerning age, gender, risk score, level of education, and cohabitation status.EQ-5D items were also imputed using EQ-5D-VAS and previous EQ-5D items.Predictive mean matching was used for the imputation of EQ-5D-VAS, nursing home days, and hours of home care services.Multinomial logit models were used for type of accommodation, level of education, cohabitation status, and EQ-5D items.The EQ-5D index was computed after the imputations.
The baseline characteristics concerning age, gender, risk score, and Charlson score were compared between the control group and intervention group.Baseline characteristics for all participants were also compared between the population in the original study and the participants in the cost-effectiveness study.We assessed differences in continuous variables using Student's t test and for categorical variables using the c 2 test.
Data were analyzed according to intention to treat.All outcomes were adjusted for age, gender, and risk score in order to correct for potential confounders.For the 10,000 simulated data sets generated using bootstrapping, adjusted mean values for costs and HRQoL for the intervention group and control group were estimated using multiple linear regression.To achieve this, the glm function, together with the EMMEANS and BOOT packages, were used in R. Remaining statistical analyses were performed in SPSS version 28 (IBM Corp).

Baseline Characteristics
In total, 369 individuals were included in the analysis.Mean age was 83.9 years, and 57% of the participants were men.We found no significant differences between intervention group or control group with regard to the basic characteristics reported in Table 1.In the original trial (1304 participants), there were significantly more women; 54% compared with 43% in this study (P < .001).No statistically significant differences were found for age, risk score, or Charlson score.

Care Utilization and Cost
The use of health care and municipal care and related costs are shown in Table 2.There were significantly fewer hospital care days in the intervention group.Costs were significantly lower in the intervention group for health care, and in total.It was only the cost of home help services that was higher in the intervention group, though not significantly higher.

HRQoL and Mortality
At the first follow-up, there was a slight but not significant decrease in EQ-5D index scores in both groups, which was maintained at the second follow-up.There was no significant difference between participants in the intervention group and those in the control group (Table 3).The proportion who died during the follow-up period was 26.2% in the control group and 24.6% in the intervention group.There was no statistical significance in mortality between the groups (mean difference 1.6%; 95% CI À0.1 to 4.1; P ¼ .23).*Unadjusted measures per group, but the differences between the groups were adjusted for age, gender, and riskscore.y The intervention is more effective and costs less than care as usual.

Cost-Effectiveness Analysis
The cost-effectiveness analysis is shown in Table 4.The difference in mean QALYs was 0.05.Care as usual was inferior to the CGA intervention in both the societal and health care perspectives, as the intervention resulted in both lower costs and gains in QALYs.The costeffectiveness perspectives in Figure 1 illustrate the uncertainty of the analysis based on the bootstrap analysis.The southeast quadrant, which implies lower costs and more effect, contains 78% of the observations, and 99% of the observations are located in the southern half of the plane, which implies lower costs.

Discussion
In this study, we found that the primary care CGA intervention is likely to be cost-effective from both health care and societal perspectives.This is mainly attributed to lower costs for both health care and municipal care in the intervention group, as the differences in QALYs derived from the EQ-5D-3L were small.
Earlier studies of primary care CGA interventions in older adults living in the community have presented conflicting evidence of costeffectiveness.Comparisons are difficult because different measures of morbidity and frailty are used, and because of differences in interventions and outcomes.A cost analysis of the GRACE intervention in Indiana published in 2009 demonstrated a lower incidence of hospitalization and emergency room visits for older adults with low income and a high risk of hospitalization in the second year of intervention. 30In the third year of follow-up, health care costs were significantly lower.An intervention from Australia was considered cost-effective at a cost of approximately USD 11,000 for reversing frailty in one older adult. 19The analysis demonstrated an effect in reducing frailty, no differences in QALYs, and higher health care costs.
In very frail subjects, the intervention was more effective and less costly.In the Netherlands, 4 well-designed interventions were performed in primary care around 2010 to 2015. 16,18,20,31None of these detected any significant differences in quality of life measures or physical functioning compared with care as usual.Three of them showed equal or higher health care costs for the interventions over 2 years of follow-up. 18,20,31The authors highlighted the heterogeneity of participants combined with challenges of recruiting participants who were frail enough as possible reasons why the anticipated effects did not appear.Furthermore, the long time that was needed for implementation of these complex interventions could result in a lag before any positive effects of the interventions could be detected.However, in the fourth (U-PROFIT) trial, health care costs were slightly lower in the intervention groups, and the intervention was found to be cost-effective at a probability of 91% and a willingness-to-pay threshold of EUR 20,000 as early as after 12 months.This 3-arm trial demonstrated the cost-effectiveness of simply identifying frail individuals in primary care and a low additional effect of a nurse-led care intervention.
The strength of the present study is that the proactive intervention was well-adapted to current practice in primary care, which may have facilitated implementation and reduced intervention costs and primary care costs.We also think that the use of a prediction model to select a sample of older adults at high risk of hospitalization allowed us to target older adults who could benefit from the intervention.We obtained reliable data concerning health care use and costs from administrative registries, with very few missing data points.Our intervention was pragmatic and adapted to the primary care context, thereby reflecting the possible effects of a broader implementation.However, costeffectiveness data should be interpreted with care outside the domestic context, as health care utilization patterns depend on local prerequisites.Although our results are in line with studies from both Europe and the United States, as mentioned previously, future studies must explore further the generalizability of our findings. 16,30here are certain weaknesses with our study.First, we could not include more than 28% of the total sample from the original study in this analysis because of informed consent.Nevertheless, the sample size is comparable to other studies, and we did not find any baseline differences between the sample in this analysis and the original study, except for a higher proportion of male participants, for which we adjusted.Second, it was not possible to randomize the practices that participated in the study.There may be differences between the practices that have influenced our results.Third, the costs for municipal care are uncertain, as the data were selfreported, resulting in large numbers of missing values, which is also the case for the HRQoL data.By using multiple imputation, data were supplemented in order to perform an analysis, but this introduces uncertainty that must be considered.We also rely on only 2 follow-up questionnaires after the baseline questionnaire and assume a stable need for municipal care until the next measuring point or death, which implies a risk for underestimation of the need for municipal care.However, we believe that this effect was similar in the 2 groups, as there was no significant difference in QALYs.Thereby, the comparison of municipal care between the groups should not be affected.

Conclusions and Implications
Our results indicate that a proactive CGA intervention in primary care for older adults with high risk of hospitalization is cost-effective under the premises of this study.The results suggest that a target group for CGA can be identified using a prediction model that uses data from medical records.It also supports the notion that a strategy for CGA with a low cost can still result in valuable effects.If the results can be reproduced, this could open up the possibility of CGA also being implemented in settings with scarce resources.

Table 1
Participant Characteristics at Baseline of the Study *Risk of hospitalization in the coming 12 months (0e1), derived from the prediction model.y Derived from medical records in the care data warehouse of the region.

Table 2
Resource Utilization and Costs During the Follow-up Period of the Study Note.Bold P values are statistically significant (P < .05).n.a, non applicable.*Unadjustedmeasures per group, but the differences between the groups were adjusted for age, gender, and riskscore.

Table 3
HRQoL Expressed as EQ-5D-Index at Baseline and During Follow-up Unadjusted measures per group, but the differences between the groups were adjusted for age, gender, and riskscore.

Table 4
Cost-effectiveness planes.Cost-effectiveness planes describing the incremental costs (y axis) and QALYs gained (x axis) from the analysis comparing the intervention with usual care in 2 perspectives.Bootstrapping with 10,000 iterations was used and incremental costs and QALYs were adjusted for age, gender, and riskscore.The upper perspective includes health care costs and the lower includes community costs together with health care costs.