Do quality improvements in assisted reproduction technology increase patient numbers in a managed competition setting?

In a system of managed competition, selective contracting and patient choice reward providers for quality improvements through increases in patient numbers and revenue. We research whether these mechanisms function as envisioned by investigating the relationship between quality improvements and patient numbers in assisted reproduction technology in the Netherlands. Success rate improvements primarily reduce volume as fewer secondary treatments are necessary, but this can be compensated by attracting new patients. Using nationwide registry data from 1996 to 2016, we find limited evidence that high-quality clinics attract new patients, and insufficiently as to compensate for the reduction in secondary treatments. The net effect of quality increases appears to be a small decline in revenue. Therefore, we conclude that patient choice and active purchasing reward quality improvements insufficiently. Nevertheless, clinics have improved quality drastically over the last years, showing that financial incentives are perhaps less important factors for quality improvements than factors such as intrinsic motivation and professional autonomy.

In a system of managed competition, selective contracting and patient choice reward providers for quality improvements through increases in patient numbers and revenue.
We research whether these mechanisms function as envisioned by investigating the relationship between quality improvements and patient numbers in assisted reproduction technology in the Netherlands. Success rate improvements primarily reduce volume as fewer secondary treatments are necessary, but this can be compensated by attracting new patients. Using nationwide registry data from 1996 to 2016, we find limited evidence that high-quality clinics attract new patients, and insufficiently as to compensate for the reduction in secondary treatments. The net effect of quality increases appears to be a small decline in revenue.
Therefore, we conclude that patient choice and active purchasing reward quality improvements insufficiently. Nevertheless, clinics have improved quality drastically over the last years, showing that financial incentives are perhaps less important factors for quality improvements than factors such as intrinsic motivation and professional autonomy.

| INTRODUCTION
Governments and health insurers aim to increase the (experienced) quality of care and the population health while containing costs, the so-called triple aim. 1,2 Quality improvements are propagated often as a means to reduce costs.
Assisted reproduction technology (ART) treatment is a good case study for this: Higher success rates lower the number of attempts and therefore lower costs and-presumably-increase patient satisfaction.
ART is an alternative when natural pregnancy fails. Treatment cycles start by ovarian hormonal stimulation, after which one or more eggs are retrieved. Eggs are then fertilized in vitro (IVF), or a sperm cell is injected directly into the egg (ICSI). In about 90% of cycles, one or two fertilized eggs are placed in the uterus. Additional fertilized eggs from the same cycle can be frozen in (cryopreservation), thawed, and placed in the uterus after a first attempt failed. This reduces the burden of starting a full new treatment cycle.
In the Netherlands, three treatment cycles are reimbursed as part of the mandatory benefit package, that is, if certain preconditions (age and body mass index) are fulfilled. Tariffs are set by the Dutch Healthcare Authority (Table 1), while volumes are freely negotiable. Success rates are defined by the percentage of 10-week pregnancies over the total number of treatment cycles. 4,5 In 2016, 13 458 treatment cycles were started, of which 6486 involved IVF and 6972 involved ICSI. An additional 12 116 embryos were placed after cryopreservation. Of the 13 458 cycles, 4781 10-week pregnancies were obtained, a success rate of 36% (Table 1).
Total success rates per started treatment cycle have increased drastically, from under 20% in 1997 to 27% in 2010 to 36% in 2016. Compared with other European countries, success rates are higher only in Sweden and Finland. 6,7 One possibility to boost success rates is to place multiple embryos at the same time. This increases the number of twins and triplets, which is a risk factor for complications. Multiplets are for that reason regarded as an undesirable outcome. However, over the past years, the percentage of multiplets dropped spectacularly in the Netherlands, from 23% of 10-week pregnancies in 2003 to 3.8% in 2016. Therefore, increases in the success rate increase did not come at the cost of increased risk of complications.
Cryopreservations add to success rates per treatment cycle, as each placement of a cryopreserved embryo is considered part of a previous treatment cycle. Increases in the use of cryopreservation will therefore increase the success rates and the total number of pregnancies. It is also cost-effective: At current tariffs, an investment of €10.000 is required to obtain a 10-week pregnancy for IVF and ICSI, while this is under €3.000 for cryopreservation.
Therefore, extended use of cryopreservation is expected to improve ART outcomes at lower costs. However, due to much lower reimbursements, higher intensity of cryopreservation also reduces per patient reimbursements for clinics.
Total reimbursements dropped from 38 million in 2010 to 34 million in 2016, which can largely be explained by increased success rates and uptake of cryopreservation.
For individual clinics, higher success rates reduce the number of treatments and thus reimbursements per patient.
Thus, a financial disincentive for quality improvements exists. On the other hand, in a competitive environment, improved success rates could attract new patients and increase total treatment volume. To attract new patients, two mechanisms coexist: patient choice and selective contracting of health insurers. 8 Preconditions for patient choice appear excellent: ART is an elective treatment, outcomes in terms of success rates are highly valued by patients, 9 and the success rates of clinics are known. 4 Due to the favorable preconditions for patient choice in AR, it can be considered a best case study. Patient choice for quality has been demonstrated in a number of studies. 8,10 Other research in the Dutch context has previously found patient responsiveness to quality indicators for angioplasty and hip replacement. 11,12 In addition, conditions for selective contracting are compelling. In 2006, Dutch health care reform introduced managed competition: insurers compete for patients under a mandatory coverage, where risk equalization prevents cherry picking. Competition on premiums incentivizes insurers to purchase actively to improve provider quality and reduce the cost of care. ART is a field in which insurers can clearly attain both of these goals: Steering patients to high-quality providers can increase quality and reduce costs. Therefore, we expect that insurers seek to steer patients through selective contracting and active purchasing strategies, especially after the 2006 reform. 8,10 For clinics, we therefore expect quality improvements to pay off in terms of revenue due to additional patient streams. 8,13 Furthermore, we expect that this effect increased after the 2006 reform. The relation may be influenced by demographics, as regional changes in the number of eligible patients (women aged 30 to 40) can influence clinic patient numbers. Also, travel time may be of influence. 14 Patients have to visit the clinic a number of times in a short time span, which could reduce willingness to travel long distances. 15 Therefore, we expect the relation between success rate and market share of the clinic to be stronger in regions where multiple clinics are within traveling distance. 10 From these premises, we can formulate the following hypotheses: 1. Changes in success rate are positively related to changes in the number of treatments.
2. This relation is stronger after the reform in 2006.
3. This relation is stronger in more competitive regions.
The paper is structured as follows: The next section presents the research method, after which our results are presented. We end with a discussion and policy recommendations.

| METHODS
To analyze the relation between success rate improvements and growth in the number of treatments, we use a panel database from 1996 to 2016, routinely collected by the Dutch Institute for Obstetrics and Geriatrics (NVOG). This institute annually sends questionnaires to the heads of department of all clinics to collect data on the number of started treatments, the number of placed embryos, the number of 10-week pregnancies, and the number of twins/triplets. Data on demographics were derived from the Dutch Statistical Bureau. 16 We define a variable (R), which is 1 when the clinic is in a competitive region, defined as the dense Randstad Area, and 0 if otherwise. 13 Success rates are published with a 1-year delay, which means that patients and insurers are expected to base their choice for the clinic in year t on quality data of year t − 1 and earlier. We test the first hypothesis by employing a fixedeffects panel regression: where T is the number of treatments in clinic i in year t. The success rate is given by σ, while D (demographics) is the number of women between 30 and 40 in the province. As a robustness check, we test different lag structures.
The coefficient β 1 contains the combined effect of two contradicting mechanisms: Increased quality primarily reduces the number of secondary treatments, while as a secondary effect, it potentially increases the number of new patients through patient choice and/or active purchasing by insurance companies. In order to test whether quality improvements attract new patients, the secondary mechanism needs to be isolated. However, our database does not contain information on individual patients. Therefore, on a patient level, no distinction can be made between first and secondary treatments. However, we are able to calculate the effect of improved success rates on the number of secondary treatments by assuming that the dropout rate after an unsuccessful attempt is fixed. In that case, the number of new treatments N is equal to where x is the dropout rate. We use a fixed dropout rate of 50%. In Germany, dropout rates between 40% and 50% are reported. 17,18 For sensitivity analysis, we range the dropout rate between 30% and 70%. To disregard general trends in the number of patients, we calculate market shares m: The regression specification is To test the influence of travel distance, we interact the coefficient with a competitive region dummy (R). 13 To test the effect of managed competition, we add interaction coefficient dummy C, which is 1 after the reform. Next, we estimate the effects of success rate on market share of new patients: In total, we estimate three specifications: (1)  As additional analyses, we include instantaneous effects and longer lag periods. Quality indicators are made available with a 1-year time lag. This means that quality in year t cannot influence patient choice or purchasing in year t.
However, instantaneous effects may occur due to reverse causality: By increasing the number of new treatments, quality of care may decline. Equivalently, quality of care may be improved by reducing the number of treatments. This effect could be caused by more strict guidelines and regulation supported by the government. 19 However, it also could potentially indicate cherry picking: By refusing patients with low chance of success, quality ratings may be improved. Thirdly, the effect could be caused by supplier-induced demand: Treating more new patients-who have lower chances of success-reduces success rates in the same year. All three effects could result in reverse causality, which could justify adding instantaneous quality effects.  Table 3 shows the main results. No significant relation is found between the number of treatments in year t and the success rate in the previous year (specification 3.1). This negative sign may signal that the effect of reduced number of secondary treatments outweighs the increase in new patients. Also, no significant effect of the success rate in the previous year on the market share of new patients is found (specification 3.2). This absence of a significant positive relation persists when interacting with region and reform (specification 3.3). We conclude that no significant relation is present between quality of care and the number of (new) treatments or the market share of the clinic.

| Main specification results
When multiple lags are tested (Table 4)  not significant. Actual competition seems to matter more than competitive reforms. This finding is robust to alternative definitions of years, dropout ratios and competitive regions (see Appendix A). In conclusion, a positive effect of quality improvements on new patients may be present, but it is insufficiently large to provide financial incentives for clinics to improve quality.   20 General physicians (GPs) acting as gatekeeper should inform patients or influence patient choice. 21 It is unknown whether GPs routinely use available quality information.
Regarding active purchasing, results indicate that at the moment, the system of regulated competition in the Netherlands does not stimulate active purchasing for assisted reproduction. As preconditions for active purchasing are better for ART than nearly any other treatment, it might be hypothesized that active purchasing by itself is falling short of policy expectations. This could be explained by low acceptance of insured patients to steering from third parties; patient choice is considered an important moral value. It could also reflect hospital market power, as resistance against selective contracting can be severe.
To improve the public's acceptance of selective contracting, insurers could increase transparency in terms of price and quality. 22 However, selective contracting is only a viable option in competitive regions. This could result in a health gap between the urban and rural regions, if predominantly urban regions are motivated to improve quality.
Indeed, our results already indicate somewhat higher success rates in the more competitive regions. Furthermore, over the last couple decades, quality has improved and costs were reduced in absence of active purchasing and despite perverse financial incentives. Our results stipulate the importance of professional autonomy of medical specialists as a means of quality improvement. The role of health insurers may be limited to stimulating professional development through, eg, trans-clinic learning programs.
This study is contributing to the literature by elucidating the effects of mechanisms of patient choice and active purchasing on incentives to increase quality in a regulated competition setting. This study has several limitations.
Firstly, as data are restricted to number of treatments, dropout rates and transfers to other clinics are unknown.
However, our results are robust to different dropout percentages in modeling. Secondly, background characteristics of patients are unknown. Some of the differences in success rates may be explained by case mix, as clinics focusing on more complicated cases may have lower success rates. However, due to the certificate-of-needs status of ART clinics, no differentiation or specialization between clinics may be expected. Clinics may also employ patient selection and cream skimming. 23 The negative correlation in the same year between success rates and patient numbers supports this. However, the large number of patients in the VU Amsterdam, which is best performing with high patient numbers, suggests that cream skimming may not be a major concern. This is supported by relatively strict guidelines for ART in the Netherlands. However, it would require additional research to definitely rule out patient selection.
To conclude, we found that patient choice and active purchasing is functioning insufficiently to reward clinics for quality improvements. Dutch ART clinics have no financial incentive for quality improvements. Despite a lack of financial stimulus, quality has improved drastically over two decades. This indicates that technological innovations and the intrinsic motivation of health professionals are the dominant mechanisms to improve the quality of care. Currently, both active purchasing and patient choice do not perform sufficiently well to ensure efficient functioning of the "market" for AR. If policy makers seek for an effective system of regulated competition, they need to either stimulate active purchasing by insurers, active choice by patients, or both.
Data used for this research is available upon request or can be accessed through the following link: https://www.

A.2 | Dropout rates
Our model assumes a 50% dropout rate. However, in literature, this figure is surrounded with a high degree of uncertainty. Therefore, we test alternative assumptions regarding the dropout rate in Table A2. We find similar results for all dropout rates, although the size of the effect declines for higher dropout rates. Therefore, we conclude that the high quality is combined with a low dropout rate, the effect is smaller. One can assume that if quality is higher, the dropout rate will be lower, suggesting that our estimates are conservative. However, the results on the number of treatments is unaffected by the dropout rate (specification one). We can therefore definitely conclude that higher success rates have no significant effect on the total number of treatments.

A.3 | Competitive region definition
Other authors have used the Randstad area as definition of a competitive region. 13 Other definitions are also possi-