A personalized intervention to prevent depression in primary care: cost-effectiveness study nested into a clustered randomized trial

Full details of the PredictD protocol and effectiveness analysis are available elsewhere [13, 14]. Briefly, the PredictD-Cluster, Controlled, Randomized Trial (CCRT) was a national, multicentre, randomized controlled trial with two parallel arms, cluster assignment by primary care centre and 18 months of follow-up (from October 2010 to July 2012).

A total of 220 primary care centres from seven Spanish cities (Barcelona, Bilbao, Granada, Jaen, Málaga, Valladolid and Zaragoza) were approached. We conducted meetings in each centre to explain the project and invite physicians to participate; 118 (53.64%) out of the 220 centres were interested in participating. Seventy centres (10 per city) out of the 118 were randomly selected. A total of 193 physicians from the 70 centres consented to participate. Of those who accepted, we randomly selected two physicians per centre (i.e. 140 physicians). Random selection was conducted using sealed opaque envelopes by an independent researcher who was not part of the research team. Randomization to intervention or control group was conducted at the centre level. In each city five centres were assigned to the control group and five centres to the intervention group.

Research assistants randomly selected four to six patients per day from the patients with an appointment with the GP, using random starting points for each day, generated using a random number generator. GPs reviewed the list each day, excluding those patients who were not eligible for the trial. A total of 8292 participants were selected. Of these, 3056 were excluded in this first stage for the following reasons: 1479 were < 18 or > 75 years old; 1039 attended the surgery on behalf of the person who had the appointment; 153 would be away (> 4 months) during the follow-up; 122 had a documented severe mental disorder; 121 did not speak or understand Spanish; 88 had cognitive impairment; 54 had terminal illnesses. The process ended when there were 26–27 eligible patients for each GP.

A total of 5236 persons were invited to participate in the study by the research assistants. Of these, 1453 patients (27.28%) declined to participate. When compared with participants, these non-participants were slightly more likely to be male (38.4% versus 36.5%) but were of similar age (50.5 versus 50.7 years).

The 3783 patients who agreed to participate were then interviewed to detect major depression using the Composite International Diagnostic Interview (CIDI). Of the 3783 patients,457 (12.08%) met criteria for major depression in the last 6 months and were consequently excluded.

A total of 3326 patients (1663 in each arm), nested in 140 GPs from 70 primary care centres, composed the sample of the trial. The total number of patients with missing data in any of the outcome variables at any point was 577 (17.35%).

Although patients did not consent to randomization, patients at the intervention centre consented to receive the intervention, and all patients agreed to data collection. Neither the patients nor the GPs were blind to the intervention, which is common for trials that evaluate psychosocial interventions [15].The interviewers who assessed outcomes, however, were masked regarding allocation to study group. Local Ethics and Human Research Committees at each city approved the protocol.

The National Health System of Spain provides universal coverage for citizens and foreign nationals (including undocumented immigrants). It is funded through taxes and free at the point of contact. Health care services are distributed into Health Areas and Basic Health Zones according to geographical, epidemiological and socio-economic criteria. Each Health Area covers a population of 200,000–400,000 inhabitants and is composed of several Basic Health Zones, which are the minimum units of health care organization. Basic Health Zones are organized around a primary care centre covering 5000–35,000 inhabitants. The primary care teams are composed of GPs, paediatricians, nurses and, in some cases, social workers. They provide a broad range of services, including the treatment of common mental disorders (shared with mental health specialists in severe cases) such as anxiety or depression [16], health promotion and preventive services. All the primary care centre staff members, including the GPs, are salaried. GP salaries contain two elements: a larger fixed payment and a smaller incentive, based on elements such as numbers of patients assigned, fulfilment of objectives, patterns of prescription and pay-for-performanceincentives [17].

The economic evaluation was conducted from two perspectives: (1) societal perspective, including the costs of all types of health services (direct costs) and the costs that stem from production losses (indirect costs), and (2) a National Health System perspective (including only direct costs from the Spanish public health services). The time frame of this study was 18 months. Therefore, we discounted both costs and effects at 3.5% following National Institute for Health and Care Excellence (NICE) recommendations [18]. All costs were expressed in euros (€) for the reference year 2012.

The participants in the two groups were similar with regard to gender (63.6% female in the control group and 63.5% in the intervention group), age (51.5 and 50 years in the control and intervention groups, respectively), marital status (68.4% and 69.9% were married in the control and intervention groups, respectively) and educational level (42.2% and 44.3% had primary level education, respectively). However, they differed in key aspects related to the trial. Participants in the intervention group had a higher risk of depression,a slightly worse mental health-related quality of life, more anxiety-related symptoms and a greater proportion of people who answered affirmatively to the two questions we use as a lifetime screen for depression [30]. In addition, there were differences in employment status, owner-occupier of an accommodation, perception of safety inside-outside the home and experiences of discrimination. Further details of trial participants are given by Bellon et al. [13].

Table 1 presents the unadjusted mean costs and effects for the intervention and control groups.

Table 2 shows the adjusted means and the ICERs.

From a societal perspective the new intervention is dominant, as the increment in cost is negative and the effects are positive. However, although most (97.4%) of the incremental effects were plotted in the Eastern quadrants (new intervention more effective), the level of uncertainty related to the costs is quite high, with half of them in the Northern quadrants (new intervention more expensive) and the other half in the Southern quadrants (new intervention less expensive). These results are depicted in Fig. 1, left column.

The acceptability curve is shown in Fig. 2. At the €30,000 per QALY ($32,058,£25,704) threshold the probability that the PredictD intervention would be seen as cost-effective was 94%. This probability increased to 98% when considering the effect in terms of the QALYs-VAS. However, these values decreased to 83% and 89%, respectively, when a threshold of €10,000 ($10,686, £8568) was used.

From a National Health System perspective, the incremental cost for QALYs gained was €1326. The cost for QALYs-VAS was €1085.45. Similarly as shown for the societal perspective, although most of the incremental effects were also plotted in the Eastern quadrants, the level of uncertainty related to the costs was quite high (Fig. 1, right column). Figure 3 depicts the acceptability curve from a National Health System perspective. Similarly, at the €30,000 per QALY threshold ($32,058,£25,704), the probability that the PredictD intervention is cost-effective was 96%, increasing to almost 100% when the effect was measured using QALYs-VAS. These values decreased to 89% and 96% when a threshold of €10,000 ($10,686, £8568) was used.

Table 3 summarizes the sensitivity analysis. The scenario that only considered the costs directly related to primary and mental health care was the best. The worst scenario was the one where the costs were doubled. However, the values were quite similar.

Over the 18-month evaluation period, the PredictD intervention was found to be efficient. The cost-effectiveness advantage arises from the finding that the PredictD intervention increases quality of life while not significantly increasing overall costs. The sensitivity analyses confirm the robustness of the results.

This is the first economic evaluation nested in a randomized trial to evaluate the effectiveness of universal prevention of depression in adults implemented by GPs. Major strengths include the large sample (more than 3000 primary care attendees) and a follow-up time of 18 months, which is longer than that for most depression prevention trials.

Nonetheless, the results of this study should be considered with the following limitations in mind. First, due to the recruitment procedure, our study may have under-represented patients who attend infrequently [31]; however, frequent attenders are more at risk of major depression [32] and are most in need of prevention. Second, intervention and control groups were unbalanced on some individual variables, so that participants had a higher risk of depression in the intervention group. This is not unusual in cluster randomized control trials, where an imbalance in the characteristics of participants can creep in because randomization occurs at the level of centre [33]. To solve that, we have adjusted the results for the unbalanced variables. Third, the patients were not blind to the intervention. They may have modified their responses to satisfy the researchers/GPs. Fourth, the information on use of services was collected by means of self-report. Some bias in recall may be expected, although it is quite likely that this bias was equally distributed between the intervention and control groups. In addition, we have not taken into account informal care-related costs and costs from general medication. As depression has an impact on physical health, it is possible that this has been affected, making the costs associated with depression possibly higher than we have calculated in our study. On the other hand, the cost associated with the training of the GP was translated to the patient level by dividing by the number of participants in the trial and not by the total number of patients, which would be more appropriate in real practice. Consequently, the costs associated with the intervention in real practice would be even lower. Fifth, in our study, only 32.4% and 36.6% of patients, in the control and the intervention group, respectively, answered yes to the two questions of lifetime screen for depression (except in the 6 months prior to the baseline interview, in which no patient suffered major depression according to the CIDI) [13]. The predictive positive value of responding yes to these two questions is 18% [30],and in our study the proportion of patients who truly suffered a first episode of depression before recruitment was 5.8% and 6.6% in the control and the intervention group, respectively. Therefore, from this point of view, our study is largely based (approximately 94% of participants) on primary prevention of the onset of depression (first episode). Lastly, the generalizability of our findings may be limited because costs associated with primary care processes in Spain are less costly than in other Western countries, due to the fact that GPs are salaried [17]. However, in these other countries, such as the USA, the cost-effectiveness threshold is also higher.

To the best of our knowledge, there are only two economic evaluations focused on the prevention of depression that can be compared with ours. Hunter et al. [34] carried out an economic modelling study based on the PredictD risk algorithm concluding that identifying non-depressed general practice attendees at highrisk of depression using the algorithm PredictD and providing them with a psychosocial preventive programme was potentially more cost-effective than the current practice. At a threshold of £25,000 (€30,000,$31,200 ) per QALY the probability of being cost-effective was around 70%. Our analysis showed that the probability of being cost-effective at this threshold is even higher.

Similarly, Van den Berg et al. [35] built up a model to estimate the cost-effectiveness of preventing depression in people with subthreshold symptoms of depression opportunistically approached in primary care practices. The intervention consisted of a self-help manual with instructions on cognitivebehaviour self-help in mood management plus up to six short telephone calls to support the participants while working through the manual. Given a willingness to pay of €30,000 ($32,058,£25,704) per disability-adjusted life year (DALY), the probability that the intervention was cost-effective was around 80%. Again, our intervention had a higher probability (96%) of being cost-effective at the same threshold. It may be hypothesized that participants in our trial benefited from a tailored face-to-face intervention as well as from the opportunistic reviews when they consulted their GPs related to other issues.

However, in both economic studies [34, 35] strategies for preventing depression in high-risk patients were evaluated (selective or indicated prevention), whereas in our study we followed a universal and personalized prevention implemented by GPs.

This intervention differs from other interventions to prevent depression because it is tailored to each patient’s individual level and profile risk. This approach parallels primary prevention of cardiovascular diseases in primary care, although the risk factors involved and their management are different. Indeed, our intervention did not increase costs because it was embedded into the day-to-day practice. Our study showed that universal prevention of depression in adults, using the PredictD intervention and implemented by GPs, has a high probability of being cost-effective compared to usual care. However, with our study we cannot know if this type of universal prevention would be more cost-effective than other types of primary prevention (selective or indicated). Further trials comparing these types of prevention and different frequencies of risk evaluation are need. GPs generally perceive that they do not have enough time to perform preventive activities, and they may agree to do so only for high-risk patients. Another possibility would be to provide universal prevention to patients by mean of a scalable and cheap strategy (e.g. through apps and smartphones), reserving the intervention of GPs only in cases of high risk of depression. Our research team is currently conducting a new trial with the latter strategy (the e-PredictD study).