Maike Schepens

10 years after their prostatectomy, the current value of the total savings in ten years is 2.8 million euros.

Volume thresholds as a stand-alone measure
Significant criticism remains for the centralization of care based on volume only. As volume is an imprecise measurement of quality, volume-based healthcare policy raises a potential risk that low-volume surgeons who provide excellent care are negatively impacted, just as high-volume surgeons who provide lower quality care are not identified. For example, the Leapfrog Group in the United States established a minimum hospital case volume of 13 for esophageal resection in a response to known improved outcomes in larger volume centers. They evaluated the variation in short-term outcomes amongst hospitals that met the volume criteria and found that although referral to high-volume centers has been an important advance for complex surgical procedures, there is still a substantial degree of variability in outcomes among hospitals. They concluded that metrics such as process, individual surgeon volume, and risk-adjusted outcome measures may yield further opportunities for quality improvement that extend beyond hospital volume-based assessments.

This supports our suggestion that a hospital volume threshold should always be accompanied by measuring outcomes, increased audit and feedback through Quality Assurance Programs (QAPs) and higher per-surgeon volume thresholds [Chapter 3].

Centralization of care: one size does not fit all
Wouters et al. studied outcomes of care before and after the introduction of a centralization project of esophagectomies for cancer and found that outcomes improved after centralization. Along with a reduction in postoperative morbidity and length of stay, mortality fell from 12% to 4% and survival improved significantly. Their study confirms that centralization of care for this type of surgery with high incidence of complications after surgery improves outcomes when patients were referred to the hospitals which showed superior outcomes in a regional audit.

However, as we concluded in our study [Chapter 3], for some interventions, simply increasing volume standards may not immediately lead to better outcomes. Despite the centralization of RPs, a striking variation in outcomes after RP remained between all hospitals, even for those with relatively high volume [Chapter 3]. As outcomes after RP were not measured, referral to hospitals with the best outcomes could not take place. In contrast to esophagectomies, direct complications after RP are limited and patients usually only stay one night in hospital after surgery. In our study we found that five-year averages of UI per hospital varied from 19% to 85%. This wide variation in UI can hardly be explained by chance variation and must be due to real differences in quality of surgical care.

We hypothesize that when outcomes are not measured, huge differences in outcomes per hospital will persist. In addition, we suggest that centralization based on volume seems insufficient to accomplish the expected improvement of quality of care. Centralization should be accompanied with continuous measurement of outcomes and quality of care improvement cycles.

Several studies examined routine outcome measurement for every patient with prostate cancer. One study investigated the integration of systematic outcome measurement into clinical practice. Compared with men followed with routine care, patients undergoing integrated quality of life assessments experienced greater recovery in sexual function scores at one year (52.2 vs 33.6; p<0.001) while no significant difference in urinary incontinence (UI) was found. However, Cathcart et al. evaluated outcomes following implementation of a quality assurance program in the UK that included monthly peer review of individual surgeon performance. They observed that patient-reported 3-months UI improved, both in terms of requirement for incontinence pads (43% before QAP and 33% after QAP; odds ratio (OR): 2.19, 95% confidence interval (CI) 1.08-4.46; p=0.02) and on International Consultation on Incontinence Questionnaire score (5.7 vs 4.2; OR: 0.82; 95% CI, 0.70–0.95; p = 0.009). For surgery where the expertise of the surgeon is the main factor influencing outcomes, such as for localized prostate cancer, improved outcomes could potentially be achieved by measuring outcomes on a per-surgeon level, by identifying best practices and by centralizing care around the best performing surgeons. Additional studies are needed to demonstrate that indeed centralizing care around the best performing RP surgeons improves outcomes on a national level. For some NQRs it is possible to analyze on a per surgeon level, with additional analyses (e.g. Dutch Heart Registration), for many other NQRs, it is not. Discussion of these results in a safe hospital setting with colleagues is of course a precarious process. We suggest that for surgeries where the outcomes are mainly surgeon-driven this feature is added to the relevant NQRs. STRUCTURE: SIGNALING PUBLIC HEALTH TRENDS IV. Trends in opioid use and prescription An opioid crisis in the US The US currently faces a serious opioid misuse epidemic that started with increased prescribing of oxycodone and the inclusion of pain as a fifth vital sign, eventually resulting in massive overdose mortality. The current addiction crisis has destroyed a multitude of lives. In the US the sales of opioids quadrupled between 1999 and 2000, and at the same time, opioid-related mortality increased from 3 per 100,000 in 1999 to 7 per 100,000 in 2010 and to 15 per 100,000 in 2017. In total 399,233 Americans died from an opioid overdose between 1999 and 2017. At the same time, various governmental agencies dedicated to solving this seemingly never-ending dilemma have not yet succeeded or delivered on their promises. Addictive behavioral seeking is a multi-faceted neurobiological and spiritually complicated phenomenon. What are the main differences between the US and Dutch healthcare system regarding opioids? In Europe, including the Netherlands, the medical use of opioids (mainly oxycodone) has also increased since 2009. Universal healthcare has been a major factor in preventing an opioid crisis of US proportions in the Netherlands for several reasons. First, all Dutch citizens are required to have a health insurance and thus Dutch people do not have to choose between high-cost care or cheaper care such as, for example, a knee replacement or chronic pain management with opioid painkillers. Second, in the Dutch healthcare system, general practitioners (GPs) are important gatekeepers to specialist care, and they integrate all patient care. GPs thereby minimize fragmentation of care. For example, 80% of opioids in the Netherlands are prescribed by a GP [Chapter 5]. In contrast, primary care physicians in the US account for only a third of all opioid prescriptions. Third, in the Netherlands public marketing by pharmaceutical companies is not allowed. Is the use of opioids in Netherlands comparable to US or European countries? Data from a report by the International Narcotics Control Board showed that in 2014-2016, the US had the highest number of Daily Defined Doses (DDDs) per million inhabitants, Germany was third in row, Belgium number seven and the Netherlands was number nine in the international row. Recent research confirmed that the US is still the number one consumer of controlled substances, with 34,731 Daily Defined Doses (DDDs) of strong opioids per million inhabitants per day. In second place is Germany, in sixth place is Belgium. Canada is in 9th, Netherlands is in 10th, UK in 21st and France in 22nd place in the world. Can claims data be used to measure trends in opioid use and prescription? We demonstrate that with just one data source, claims data, relevant information for healthcare professionals and policymakers about opioid use and prescription can be acquired [Chapter 5]. For each Dutch citizen we selected claims data for all opioids, except codeine and buprenorphine, for the period 2010-2017. A total of 3,655,265 different insured persons used opioids during the research period. The yearly number of opioid users increased from 650,864 in 2010 to 1,010,474 in 2017 [Chapter 5]. This increase was mainly driven by an increase in oxycodone prescriptions. Chenaf et al. studied opioid use based on claims data in the French population and also found that opioid prescriptions at least doubled in the period 2004 until 2017 and that oxycodone use increased particularly. We found that elderly and female patients most frequently used opioids. These findings are also in line with international literature. The ratio of short- versus long-term opioid users remained steady during the research period, with opioids being used for four months or longer in 21% of cases. General practitioners prescribed the largest share of opioids, but a growing number of prescriptions originated from medical specialists [Chapter 5]. Compared to use of claims data only, more insight was gained by performing a multi-source database study. Kalkman et al used a combination of national registries to explore opioid prescriptions and several proxies for misuse, including addiction, hospitalizations, and mortality. Their study demonstrates how much insight can be gained with multi-source databases, even without connecting data sources. Their findings clearly show an increase in opioid prescriptions being paralleled by an increase in multiple proxies for opioid misuse. Compared with the US however, the use and misuse of prescription opioids and opioid-related mortality were still very low. Future research on signaling health trends Claims data can be used to explore trends on a national level for a plethora of healthcare domains. For example, a recent nationwide study on chemotherapy use and intensive care unit (ICU) admission in the last three months before death in patients with cancer of the stomach or esophagus was also based on claims data only. The study found that chemotherapy use and ICU admission shortly before death were relatively infrequent in the Netherlands. Chemotherapy was used less often in hospitals that treat many patients compared to hospitals that treat fewer patients. Comparable studies could be executed on the use of expensive drugs in the last months of life, for example for several oncological conditions. PROCESS: OPTIMIZING INEFFICIENT PROCESSES V. Using existing data for national quality registries The administrative burden of national quality registries As stated before, centralization based on volume only seems insufficient to accomplish immediate improvement in quality of care. Centralization should be accompanied with continuous measurement of outcomes and quality improvement cycles and more time might be required to establish effect on a national level. NQRs are the perfect tools for these measurements, however they still have one big disadvantage: Having to perform double registrations due to shortcomings in digital systems is perceived as a barrier for NQRs. To date, there is very limited research on how the administrative burden for NQRs can be reduced on a national level. In a recent study, Zegers et al evaluated time spent on quality administration for three large hospitals and five different departments and care trajectories and found that the average Dutch healthcare professional spends 52.3 minutes a day on administration in the context of accountability of the quality of care both in electronic health records (EHRs) and other databases. These quality data are requested by government bodies, accreditation institutes, insurers, professional associations, patient organizations and hospital boards. The average number of quality measures per department is 91, with 1,380 underlying variables. Only 25% of these data is required for quality improvement. The administrative burden on the clinical level may not only reflect operational inefficiencies, but also failures in governance at macro- and meso level. The impossibility of exchanging data between hospitals with different EHR systems and the administrative burden of registration both should be more firmly on the policy agenda. Where Zegers et al. do plea for less quality registries, a limited set of core indicators and a better use of information and communication technologies to reduce these workloads, we demonstrate another possibility, without losing the full potential of the impact of data from NQRs to use in quality improvement cycles. What is a clinical information model and how can it be used? Clinical information models (CIMs) can be seen as building blocks collecting different data elements. They are needed for multiple reuse of data and were first introduced in the Netherlands around 2010. Figure 1 describes the structure of a clinical information model (CIM). Figure 1 Structure of a clinical information model CIMS can be used for many different purposes (figure 2). At the moment, Dutch programs for all these purposes each have their requests for adaptation of specific CIMs and/or the overall CIM-structure (figure 2). Figure 2 Purposes for use of data from electronic health records structured with clinical information Models and examples of Dutch national programs Can existing clinical information models be used for capturing data elements for NQRs? The potential of using existing CIMs (also called clinical building blocks) in EHR systems for data collection for national quality registries (NQRs) is high. The average percentage of data elements for NQRs that can be captured from EHR systems by using existing CIMs is 83% [Chapter 6]. To our knowledge, this is the first study which matches data required for NQRs with CIMs. Matching of these data elements is the first step in exploring the potential. Implementing CIMs in hospitals and reusing the EHR-data for NQRs will be the next step. Unfortunately, there is very limited international scientific literature on the subject of the potential and the implementation of CIMs. This may be due to the fact that hospitals and/or regions design and implement their own solutions for data reuse and do not translate this into a scientific contribution. Future perspectives In theory EHR data can be used to reduce the administrative burden for NQRs as we demonstrated a high potential coverage of data elements of NQRs with existing CIMs [Chapter 6]). Yet, this is only possible when CIMs are implemented nationwide in EHRs and in systems of NQRs. In order to use EHR data structured with CIMs for NQRs, several implementation steps need to be taken, such as: 1) Compliance to CIM-structures and codelists for EHR-systems and NQRs 2) Focus of different national programs should proceed from perspective of single reuse of data to multiple reuse of data (Collect Once Use Many Times (COUMT)) 3) Healthcare professionals should make a transition to more structured and standardized documentation Each of these steps will be explained/ illustrated below. Ad 1) Compliance to CIM-structures and codelists To support adequate implementation of CIMs, we checked a few examples on the level of compliance across different EHR-systems for some CIM structures and corresponding codelists (figure 3). Figure 3 Level of compliance to clinical information models (CIMs) of three electronic health record (EHR) systems for five CIMs This analysis demonstrates that to date, the main EHR-systems seem not yet to be compliant with the current CIM-structure, and corresponding codelists. A more detailed analysis for one codelist clarified even more what the current situation is. We previously studied compliance to the codelist ‘Tobacco use status’ of the data element ‘Tobacco use status’ of the CIM ‘Tobacco use’. We found that three different EHR-systems use three different sets of data elements. In total only two out of seven codes were implemented compliant to the codelist in three EHR systems, and only one of the EHR-systems was fully compliant with the codelist (figure 4). Figure 4 Example of compliance to clinical information model (CIM) on detailed level for codelist ‘Tobacco use status’ for CIM ‘Tobacco use’ for three Electronic Health Record Systems Another example of content standardization is the CIM ‘Operations’ with the related codelist ‘Operations thesaurus’. To date, this codelist is implemented in only a few hospitals in the Netherlands. Thus, the implementation of one of the five most important CIMs should go hand in hand with a national implementation of this related codelist. Content standardization is key priority in the Netherlands and requires national governance. Ad 2) Shift of focus from single to multiple reuse of data Data is a major asset that should be considered as strategic for any clinical organization and essential for every healthcare professional. Reuse of clinical data is crucial for healthcare quality, management, reduced costs, population health management and effective clinical research. However, most research demonstrates that possible advantages of clinical data reuse still lay in our future. In the Netherlands many national programs currently focus on single reuse of data for a specific purpose, which might contribute to the current hick-ups in national implementation of CIMs. A common goal for multiple reuse of data following the COUMT paradigm, might also substantially contribute to the alignment and cooperation of the different national programs (figure 5). Working together with all programs towards a national goal, may optically slow down results of a single program, but is likely to eventually lead to improved outcomes and reduced costs for all programs. Figure 5 Proposed shift of focus from goal per national program to common goal and multiple reuse of data Without content standardization and a shift of focus to multiple reuse, following the Collect Once Use Many Times (COUMT) principle is impossible and exchange of data will remain limited to hospitals using the same EHR system. In the Netherlands, many regions have hospitals using different EHR systems, therefore data exchange is hampered. Ad 3) Transition to more structured and standardized documentation The primary purpose of clinical documentation is to support high-quality patient care. The results of a retrospective multicenter study showed that structured documentation is associated with higher quality documentation, with a 20% increase in documentation quality measured on a 0–100 scale. There could be a concern that as data reuse becomes more important, healthcare providers are required to capture even more data while providing care. This, in turn, might increase the administrative burden. This should be avoided at any cost, as healthcare providers are unlikely to accept a documentation method that adds a significant burden to their workload. Efforts should be made to implement structured documentation methods within EHRs to enable data reuse while reducing the administrative burden. The pandemic opened up many windows of opportunity for positive reforms, and now may be the time to address this important digital transition in healthcare in a fundamental way on an (inter)national level. VI. Conclusions The overall aim of this thesis was to contribute to the body of knowledge whether it is possible and useful to measure and improve quality of healthcare by using secondary data, such as claims data. This thesis shows that claims data can indeed be used to measure outcomes of care, to evaluate quality of care by quality improvement cycles and to evaluate trends in healthcare on a national and local level. Even volume-outcome relationships for certain procedures can be studied by using claims data. A wide variety of patient outcomes is seen in hospitals for lower disk hernia surgery and radical prostatectomy (RP). For RP there is a clear volume-outcome (urinary incontinence (UI)) relationship, yet even within high-volume group of hospitals there is a wide variation in outcomes. Patients operated in hospitals that increased the volume of RPs over time, had a 29% lower risk of UI than patients operated in hospitals that remained of low volume (120 RP per year). Patients operated in hospitals that remained high volume (>120 RP per year), had a 52% lower risk of UI than patients operated in hospitals that remained of low volume.

Volume thresholds without measuring outcomes seems to be insufficient to improve quality of care within a few years after increasing the volume threshold. Measuring of outcomes is necessary on a national, hospital and sometimes even per surgeon level. Centralization of care is not a one size fits all. For procedures in which the expertise of the surgeon is one of the main determinants of outcome, such as for RP, centralization should be accompanied with proceeding specialization, expressed in the number of procedures performed per surgeon. So far, centralization of RP has taken place in the Netherlands, yet the same number of urologists performed these RPs.

National quality registries are a great source of information for registering healthcare outcomes and improving quality, however the perceived and actual administrative burden is high. Reusing data from Electronic Health Records, structured with clinical information models (CIMs) has a high potential for reuse of data: 83% of required data for more than 30 national quality registries can be based on existing CIMs.

The knowledge that stems from this thesis, can be transferred to other areas and other diseases, and in this way contribute to improving outcomes for patients. Transparency of hospital-specific outcome information is a prerequisite for the continuous process of quality improvement and it is a legal right for patients to be informed about differences in outcomes per hospital.

Recommendations
The Dutch Integral Healthcare Agreement (Integraal Zorg Akkoord [IZA]) has several ambitious goals for the coming years. Based on this thesis, several recommendations are made which will support these IZA goals and benefit patients:

Appropriate care:
- Development and implementation of NQRs, that include standard sets of outcomes that matter to patients, at least for those conditions with significant health burden and/or societal impact such as prostate cancer and lumbar disk herniation.
- Reconsider the strict interpretation of the GDPR and make claims data available for scientific research, more specifically the study of outcomes of care.
- Make use of existing data, such as claims data to evaluate outcomes of care for more procedures in order to reduce the administrative burden for healthcare professionals.

Centralization of care:
- Centralization of specific care can indeed increase the quality of care. When adopting centralization however, this should always be based on a scientific analysis of its effects. Centralization should not be a goal in itself, the goal is to improve outcomes.
- When volume thresholds are installed, it should be accompanied with measuring of outcomes, and not as a stand-alone measure to improve and/or indicate quality of care.
- Centralization of care should be adapted to the specific procedure; procedures with high complexity in surgery only (such as radical prostatectomy), should be centralized around the best surgeons; (procedures with high complexity and high risk after the procedure, should be centralized around the best teams).

Electronic exchange of data:
- The five most used clinical information models (CIMs) should be implemented in all healthcare domains in the Netherlands while following the COUMT-paradigm. The code- and value lists (such as the Operations Thesaurus) related to these five CIMs should also be implemented nationwide. Implementation could start in hospitals and NQRs.
- Adherence to (inter)national codelists is a sine qua non for national implementation of CIMs and reuse of data.
- All national programs for data reuse should be in alignment with the COUMT paradigm, more specifically a focus on multiple reuse of data for different purposes.