In this study we have investigated the short-term changes in costs and effects after the implementation of 16 DMPs for three different chronic diseases, namely CVR, COPD, and DMII. We have also explored the within DMP predictors of these changes. Finally, a CUA was performed from the health care and societal perspective comparing each DMP to usual care and comparing the most effective and least effective DMP within five disease categories (i.e. CVR-primary prevention, CVR-secondary prevention, CVR-both types of prevention, COPD, DMII).
Our results show a significant improvement in the level of chronic care integration as measured by the PACIC, in the CVR population (0.10). It improved especially in the DMPs that were directed at primary prevention (0.18) or the combination of primary and secondary prevention (0.10) of cardiovascular diseases. This is promising because patients in these programs had the lowest PACIC scores of the three patient groups. For patients who already had a cardiovascular disease it is probably harder to achieve improvements in integrating care because more (para-) medical disciplines and healthcare sectors become involved. An unexpected result was that the PACIC decreased by 0.23 in the DMII-DMPs. This may be due to difficulties to maintain their high starting level of integrated care, which in turn may be caused by the attention that was paid to quality improvements in diabetes care for the last decade. It would be interesting to examine whether our findings would have been similar if another instrument, for example the Assessment of Chronic Illness Care (ACIC), would have been used to measure the level of chronic care integration. However, we did not include the ACIC in our analysis for two reasons. The first is because this paper focuses on intermediate and final outcomes in patients, not in professionals. The second is that although the two instruments are complementary , they both measure the level of integrated care and thus, they correlate .
Another interesting finding is that DMPs seem to improve the life-style of patients, in all three disease categories. Patients reported a higher level of physical activity, especially those in DMPs for COPD and CVR management. In addition, the percentage of smokers decreased by more than 5 percentage-point in all disease categories; the decrease was 11 percentage-point in COPD. This reduction is considerably higher as the cessation rate achieved by a physician-advice to stop smoking  or the impact of the recent ban on smoking in bars and restaurants .
Furthermore, our within-DMP analysis showed a reduction in self-efficacy and generic HR-QoL after the implementation of the DMPs. The slight deterioration (about 0.03 EQ-5D units) in HR-QoL may be explained as a time effect rather than a treatment effect because the HR-QoL of chronic care patients generally tends to decrease over time . Similarly, the decrease in self-efficacy may also be related to the decrease of HR-QoL because deterioration in HR-QoL may worsen self-efficacy [24, 25]. Another explanation may be that HR-QoL and self-efficacy are both perceived values that are influenced by the information and knowledge a patient has. DMP interventions included educating patients about their disease, learning them to recognize the early signals of disease-worsening, learning them coping skills and stimulating them to improve their lifestyle. As a result, patients may have become more aware of their impaired health status and their reference point may have shifted.
Our study collected the costs of development and implementation of the DMPs in detail and showed that they can be an important driver of total costs. This is in line with the findings of the few previous studies that have incorporated them in their analysis [3, 26, 27]. The development and implementation costs per patient were largely driven by the personnel costs. Moreover, the 16 DMPs included in our sample were pioneers in experimenting with DMPs. Therefore, the number of enrolled patients was perhaps not as high in the first year of implementation as the capacity would allow. In the long(er) term, we expect that more patients will be enrolled in the DMPs and caregivers will gain experience in managing and maintaining a DMP. That may lower the implementation costs per patient. Therefore, we would expect more favourable ICERs for the DMPs in the longer term. Within the one-year time frame of our study there are as yet few signals of important changes in the costs of healthcare utilization and productivity loss. But the heterogeneity in DMPs is large with all 3 DMII-DMPs showing a numerical reduction of hospital costs and total health care costs.
The regression analysis indicated that an increase in physical activity was predictive of an increase in HR-QoL. Given the observed increase in physical activity in almost all disease categories, we may expect DMPs to improve HR-QoL in the longer term. We also found that an improvement in self-efficacy was predictive of an improvement in HR-QoL. This creates an opportunity for DMPs to develop and implement strategies to improve the self-efficacy of the patients. Furthermore, patients with multiple morbidities seem to benefit less than patients with one disease. This may imply that the current disease-specific DMPs do not address the needs that patients with multi-morbidity have, and therefore, are less effective for this population. The need for patient-tailored care to address the complex needs of patients with multi-morbidity is extensively addressed in the literature [28, 29]. A horizontal integration of DMPs to simultaneously target CVR, COPD, and DMII might be appealing for several reasons. The first one is of course the desire to improve the care for these patients. The second reason is that some components of the DMPs are largely similar, irrespective of the disease. For example, smoking cessation support and physical reactivation can be organized similarly, and adjusted to the specific needs of an individual patient. This avoids inefficiencies and double payments. Another reason is that the number of participants in such a multi-disease DMPs will increase, which will lower the implementation and overhead costs per participant.
We also performed a CUA comparing DMPs within a disease area, which is interesting for decision makers once they have decided to implement a DMP. Then the variability in costs and health outcomes is likely to drive the choice of program. When adopting the health care perspective the CUA showed that the majority of the bootstrapped ICERs in all types of CVR prevention and DMII comparison pairs were located on the South-East quadrant of the CE plane. This indicates that the most effective DMPs had lower costs and positive QALY gains compared to the least effective DMPs in these three disease groups. This finding remained also when the societal perspective was adopted. However, the results concerning the primary CVR prevention and COPD were more difficult to interpret because of the uncertainty about the QALY gains (health care and societal perspective).
As our results showed, the cost-effectiveness of DMPs varies considerably, most likely depending on the components of the program, the target population, the success of the implementation and the costs of managing and operating the program. These are all factors that contractors of DMPs should consider in the negotiation phase. We are planning future analysis aiming to identify the factors that drive the cost-effectiveness of a DMP. These findings could contribute to the on-going debate in the Netherlands on whether the current bundled payment system for single-disease DMPs are an intermediate stage towards population-based financing . Population based financing includes a risk-adjusted fixed budget (either per group of patients or region) to cover all health care provided by multiple professionals from different disciplines. Savings compared to a pre-defined benchmark are often shared between payer and provider. A large variation in the cost-effectiveness of DMPs due to the aforementioned factors, jeopardizes the successful implementation of DMPs as means to achieve integration of chronic care. Thus, a population-based financing with larger scope in terms of covered population and provided interventions, economies of scale that lower operating costs, and consensus of all stakeholders that ensures successful implementation may appear attractive to Dutch policy makers. However, the preconditions to introduce a population-based financing are far from being reached  and therefore, the implementation of DMPs on more disease areas is still work in progress.
This study contributes to the growing body of international evidence on integrated care in several ways. First, it highlights the necessity to adopt a broad set of outcome measures and include the most important cost items from different perspectives in the evaluation of DMPs. Second, the findings of our study support the previous studies that concluded that DMPs are positively associated with improvements in patient lifestyle and quality of care [20, 31, 32]. Third, our finding that DMPs have the potential to become cost-effective in the long-term, and the identification of factors that drive that cost-effectiveness, could inform designers of integrated care programs in other European countries. Fourth, the limitation of disease-specific DMPs to address the needs of complex patients could urge collective initiatives on a European level to develop adequate models of integrated care for this population.
Our study is one of very few studies providing insight into health economic aspects of DMPs that includes such a broad range of outcome measures and cost categories. However, we fully acknowledge the limitations of the study design with respect to causality. At the start of this study there were multiple initiatives to provide integrated care across the entire country, stimulated by the introduction of the bundled payment system and other financial incentives. Therefore it was impossible to create a control group at regional level. It was also difficult to identify control groups within the same organization because of the high risk of contamination . This risk is high because the implementation of a DMP requires changes at an organisational level. For example, redesigning the care-delivery process or training nurses in motivational interviewing affects the entire organisation and the entire target population. Therefore, we did not aim to compare the DMPs to usual care but rather compare different DMPs within a disease category. To optimize comparability, we applied inverse probability weighting and corrected for confounders in multivariate analysis. In addition, our results may be object to regression to the mean bias. However, this bias is probably limited because our sample size is relatively large and the diseases included in our analysis are chronic and progressive. These assumptions are supported by a previous study that found minimal evidence of regression to the mean in COPD-DMPs .