Economic evaluation of stepped-care versus usual care for depression and anxiety in older adults with vision impairment: randomized controlled trial

An economic evaluation based on a societal perspective was performed alongside a two-armed multicenter randomized controlled trial (RCT), as described in the original protocol [22]. The trial is registered in the Dutch trial registry (identifier: NTR3296, http://​www.​trialregister.​nl). Detailed information on the study design and intervention is provided elsewhere [21, 22].

Patients were eligible if: a) they had subthreshold depression and/or anxiety, i.e. scored ≥16 on the Centre for Epidemiologic Studies Depression scale (CES-D) [23, 24] and/or ≥8 on the Hospital Anxiety and Depression Scale-Anxiety subscale (HADS-A) [25, 26], b) they did not meet the DSM-IV criteria of a major depressive, dysthymic and/or anxiety disorder as assessed with the Mini International Neuropsychiatric Interview (MINI) [27, 28], c) they adequately spoke the Dutch language, and d) they were not severely cognitively impaired, as assessed with the Six-item screener [29].

A pre-specified power calculation was based on the study of van ‘t Veer et al. [30] who evaluated the cost-effectiveness of stepped-care in the general elderly population. They found an effect size of 0.44. Based on a two-sided α ≤ 0.05, a power of 0.85, and a dropout rate of 20%, a minimum of 230 patients (115 in each arm) was needed. However, since we observed higher dropout rates than expected at the start of the RCT, more patients were included (n = 265).

Participants were randomly assigned to stepped-care plus usual care (n = 131) or usual care alone (n = 134). An allocation scheme was generated by a computerized random number generator stratified by outpatient clinic and based on blocks of two. After the baseline measurement, an independent researcher performed randomization, which was registered at the low vision rehabilitation center. From the beginning of September 2012 to the end of July 2015, seven telephone interviews (baseline, and after 3, 6, 9, 12, 18 and 24 months) per participant were performed at the VU University Medical Centre in Amsterdam by trained and masked research assistants. Participants were told not to divulge the nature of their treatment. Due to the nature of the stepped-care intervention, therapists and patients could not be masked.

The stepped-care program consisted of four steps that each lasted approximately 3 months The total intervention period lasted 12 months, followed by a 12-month follow-up.

At the end of each 3-month period, elevated levels of depression and/or anxiety (i.e. ≥16 on the CES-D and/or ≥8 on the HADS-A) induced moving on to the following step of the intervention. The first step was a period of watchful waiting, which involved an active decision to not directly treat the depressive and/or anxious symptoms but, instead, to intermittently re-assess these symptoms. In the second step, a guided self-help course based on cognitive behavioral therapy (CBT) was given. The course was available in written, digital, audio and Braille formats and was supported by trained occupational therapists. In the third step, problem solving treatment (PST) was offered by trained social workers and psychologists. Finally, when participants still had subthreshold depression and/or anxiety after the third step, they moved to the fourth step, which was a referral to their general practitioner (GP) to discuss other treatment and the use of medication. Participants who developed an actual depressive and/or anxiety disorder as assessed with the MINI, were directly referred to their GP. Usual care for both the stepped-care and usual care group included low vision rehabilitation care and/or care that was offered by other healthcare providers.

The primary outcomes were QALYs and the cumulative incidence of major depressive, dysthymic and/or anxiety disorders (i.e. panic disorder (without agoraphobia), agoraphobia (without a history of panic disorder), social phobia, and/or generalized anxiety disorder). The latter was assessed at every measurement time point with the Dutch MINI Plus (5.0.0) [27, 28]. QALYs were determined by measuring health-related quality-of-life at baseline and at 12 and 24 months using the EuroQol (EQ-5D-3 L) using the official telephone script; this instrument comprises five dimensions (mobility, self-care, activities of daily living, pain/discomfort and depression/anxiety) with three response options (no problems, some problems, extreme problems) [31]. The EQ-5D-3 L health states were converted to health utility scores using the Dutch tariff, in which 0 corresponds to death and 1 corresponds to full health (range − 0.33 to 1, whereby negative utilities indicate that a health state is valued as worse than death) [31]. With the area under the curve method, QALYs were calculated by multiplying the amount of time a patient spent in a particular health state by the utilities. Changes in utilities between health states were considered to be linear. Secondary outcomes were change in symptoms of depression and anxiety, as assessed with the CES-D and the HADS-A, respectively. The CES-D has 20 items and the HADS-A has 7 items evaluated on a 4-point Likert-scale [23‐26]. In the cost-effectiveness analyses, the change of symptoms between the start and end of follow-up was assessed instead of the course of symptoms over time as was done in the effectiveness analysis [21]. This was done to facilitate the interpretation of the incremental cost-effectiveness ratios.

Costs were collected from a societal perspective (informal care was not included). Healthcare utilization was measured using the self-rated Trimbos and iMTA questionnaire for Costs associated with Psychiatric illness (TiC-P, adapted version) [32] and valued using standard costs from the Dutch costing guideline (Table 1) [33]. Medication was valued using prices from the Royal Dutch Society for Pharmacy. Lost productivity due to absenteeism from paid and unpaid work and presenteeism were measured using the Short Form Health and Labour Questionnaire (SF-HLQ) [34]. Costs of absenteeism from paid work and presenteeism were calculated using mean age and gender-specific income values of the Dutch population and calculated according to the friction method. Compared to the human capital method which assumes any hour not worked as an hour lost, the friction method assumes that after a certain period of time (i.e. 161 days) the sick employee is replaced [35]. Thus, lost productivity costs are generated only during the friction period. Lost productivity costs from unpaid work were valued using a shadow price for informal care (€13.50/h) [33]. Both the TiC-P and SF-HLQ were administered at 6,12,18 and 24 months follow-up with a recall period of 6 months. The costs of the stepped-care program were calculated using a bottom-up approach, i.e. costs were determined by the resources each person used at each step (time, materials, etcetera). Standard costs for occupational therapists, social workers and psychologists obtained from the Dutch costing guideline were used to value these resources (see Table 1). The index year was 2013. If necessary, consumer price indices were used to correct prices [36].

First, non-response analyses, dropout analyses, and comparisons of baseline differences between the stepped-care group and usual care group were performed with χ²-tests, independent samples t-tests, and non-parametric tests. Second, a cost-effectiveness and cost-utility analysis were performed based on the intention-to-treat principle.

The effectiveness analyses are reported elsewhere [21]. In contrast with the effectiveness analyses, in the present study missing cost and effect data were replaced using multiple imputation with chained equations (MICE) [37]. The number of imputed datasets was increased until the loss of efficiency was less than 5% resulting in 15 imputed datasets [38]. Variables in the imputation model included all outcome variables, characteristics differing between groups at baseline, variables related to missing data and variables related to the outcome variables. To account for the non-normal distribution of cost data, predictive mean matching was used in the MICE procedure [38]. Results from the multiple datasets were pooled using Rubin’s rules [39]. Bivariate regression models were used to estimate cost and effect differences, and incremental cost-effectiveness ratios (ICERs) were calculated. Bias-corrected and accelerated bootstrapping was applied to estimate 95% confidence intervals (CI) around the mean cost and effect differences (5000 replications). The bootstrapped cost-effect pairs were plotted on a cost-effectiveness (CE) plane for each outcome separately to show the uncertainty around the ICER. The net benefit approach was used to estimate a cost-effectiveness acceptability curve (CEAC) using the pooled cost and effect differences, and the pooled, bootstrapped standard errors. The CEAC shows the probability that stepped-care is cost-effective in comparison with usual care for a range of different ceiling ratios (i.e. the willingness-to-pay for one additional recovered patient), indicating decision uncertainty [40].

For the analyses, SPSS for Windows version 21 (SPSS IBM, New York, USA) and Stata/SE software, version 12 (Stata Corp LP) were used.

The primary analysis was based on costs from a societal perspective using the friction method to estimate indirect non-healthcare costs.

In addition, two sensitivity analyses were performed: i) the cost-effectiveness analysis was performed from a healthcare perspective, i.e. including only direct costs, ii) also, to determine productivity losses, the human capital method was used in which every hour not worked is considered as a lost hour.

Of the 3000 invited participants, responders (n = 914; 30%) were significantly younger than non-responders (n = 2086; 70%) (mean difference 4.6 years, p < 0.001). At 24-month follow-up, of the 265 patients eligible/willing to participate, 91 dropped-out (34.3%; 45 stepped-care and 46 usual care). Participants who dropped-out of the study more often lived in a nursing home and were significantly older than those who did not drop-out (p < 0.05). Common reasons for dropout were: mortality, physical or mental inability to continue, or too heavy a burden.

Table 2 presents the baseline characteristics of the two study groups: there was a significant difference in education level between the groups (p < 0.05).

The cumulative incidence of depressive/anxiety disorders at 24-month follow-up was 0.29 in the stepped-care group and 0.46 in the usual care group. The absolute risk reduction was 0.17; this difference was significant (95% CI 0.06 to 0.29) (Table 3). In addition, imputed and pooled outcomes showed a significant difference between the stepped-care group and usual care group for the HADS-A (mean difference 1.43, 95% CI 0.10 to 2.77), and a non-significant difference for the CES-D (mean difference 2.73, 95% CI -0.28 to 5.74) and QALYs (mean difference 0.03, 95% CI -0.09 to 0.15). Note that these latter analyses are different from those performed in our earlier study.²¹

Cost and effect data are presented in Table 3. The mean total intervention costs amounted to €262 per participant. Although direct healthcare costs were lower for the stepped-care group compared with the usual care group, the mean difference was not significant (−€1154; 95% CI -7708 to 4328). Cost savings were mainly due to significantly lower secondary mental healthcare and hospitalization costs. Indirect non-healthcare costs based on the friction method were not significantly higher for the stepped-care group compared with the usual care group (€277; 95% CI -1418 to 2230). Total costs were lower for the stepped-care group compared to usual care (−€877; 95% CI -8039 to 5489) but the difference was not significant.

The cost-utility analysis resulted in an ICER of −29,233 indicating that to gain 1 QALY €29,233 is saved in the stepped-care group as compared to usual care. The CE plane and the CEAC in Fig. 1 a and b show that the probability that stepped-care was cost-effective compared to usual care was 59% or more for a ceiling ratio of €0 per QALY and that this increased to 65% or more at a willingness-to-pay of €20,000 per QALY.

With regard to the cumulative incidence of depressive/anxiety disorders, the ICER was −5159 indicating that to prevent one case of depression or anxiety €5159 is saved in the stepped-care group as compared to usual care (Table 4, Fig. 1). The CE plane and the CEAC in Fig. 1 c and d show that, for a ceiling ratio of €0 per disorder prevented, the probability that stepped-care was cost-effective compared to usual care was 59%. At a willingness-to-pay of €10,000 this probability was 77%, and at a willingness-to-pay of €20,000 this probability was 88%. The probability of cost-effectiveness increased to 95% or more at a willingness-to-pay of €33,000 per disorder prevented.

For the CES-D and the HADS-A, cost-effectiveness acceptability curves show that the probability of cost-effectiveness was 60% for both outcomes for a ceiling ratio of €0 per point improvement on the CES-D/HADS-A; this probability increased to 95% or more at a willingness-to-pay of €2500 per point improvement on the CES-D and €4000 per point improvement on the HADS-A (Table 4).

Based on the societal perspective with the human capital method (non-significant) higher total costs were found for the stepped-care group compared to the usual care group (mean difference 200, 95% CI -7035 to 6829) (Table 4). The cost-effectiveness acceptability curves show that the probability of cost-effectiveness was 47% or more for a ceiling ratio of €0 per disorder prevented and that this increased to 95% or more at a willingness-to-pay of €42,500 per disorder prevented. Based on the healthcare perspective, compared with the main analysis a larger cost difference was found, with stepped-care being less costly than usual care (mean difference − 1154, 95% CI -7708 to 4328) (Table 4). The cost-effectiveness acceptability curves show that the probability of cost-effectiveness was 59% or more for a ceiling ratio of €0 per disorder prevented and that this increased to 95% or more at a willingness-to-pay of €26,000 per disorder prevented (see Additional file 1: Figure S1 with CE-planes and CEACs).

A strength of this study is the pragmatic design that was chosen. This helps policymakers make evidence-based decisions on whether implementation of stepped-care can be considered an efficient allocation of scarce resources in real-life situations and to increase generalizability of the results. Second, a long follow-up period was chosen to assess long-term treatment effects. Third, treatment arms appeared to be well randomized with no relevant differences in baseline characteristics between the stepped-care and usual care group. A limitation of this study is the high dropout rate: 34% in total. However, there was no significant difference in dropout between the stepped-care and usual care group and there was sufficient statistical power. Missing data were dealt with by applying multiple imputation techniques, which is the preferred method in economic evaluations [43]. Second, the relatively large recall periods for healthcare utilization and work productivity (6 months) may have introduced recall bias. Third, informal care was not included in the total costs from a societal perspective. However, it is hard to estimate how inclusion of informal care costs would have influenced our societal cost estimates. Fourth, although the pragmatic design of our study increases generalizability and the outcomes could be used for stepped-care in different settings, strictly speaking our outcomes can only be generalized to visually impaired older adults who are registered at a low vision rehabilitation organization.

This study shows that stepped-care is dominant to usual care (with a probability of around 60%) in treating mental health problems in visually impaired older adults. Stepped-care enables professionals to efficiently deploy their limited resources by initially offering low intensity and low cost interventions and only moving on to higher-intensity and more costly interventions when sufficient response is lacking. The stepped-care program is effective in preventing major depressive/anxiety disorders as compared to usual care; the probability that stepped-care is cost-effective compared to usual care is 95% or more at a willingness-to-pay of €33,000 per disorder prevented. Future studies should investigate how the cost-effectiveness of stepped-care can be improved. Possible options for this are: offering interventions tailored to personal needs and symptom severity, e.g. varying the watchful waiting period [44], and directly offering higher intensity interventions to persons with a history of depressive/anxiety disorder [21]. In addition, other evidence-based treatment options (e.g. exercise programs, e-mental health) could be added to the model. Moreover, the present study, which investigates both the clinical effectiveness and cost-effectiveness of a promising intervention, could serve as an example for other intervention studies in the field of low vision. Considering that vision impairment is ubiquitous in an aging population and that resources in healthcare are scarce, such economic evaluations are highly relevant.