Examining the Long-term Spillover Effects of a Pay-forPerformance Program in a Healthcare System That Lacks Referral Arrangements

Background

Over the past two decades, an increasing number of pay-for-performance (P4P) programs, which use financial incentives to better motivate physicians and improve the quality of healthcare, have been adopted worldwide.^1-5 A P4P program is expected to enhance the quality of care by offering incentives to providers. Financial incentives reward certain services based on pre-determined benchmarks. These programs generally provide incentives to healthcare providers to improve the process of care (eg, checking the glycated hemoglobin [HbA1c] levels of patients with diabetes) and intermediate outcomes (eg, controlling the HbA1c levels of patients with diabetes). The measures are based on the recommendations of clinical practice guidelines. Good adherence to clinical guidelines can improve health outcomes for patients. Therefore, considering the incentives offered under P4P programs, healthcare providers might devote more effort to specific conditions or indicators that are rewarded and pursue better care outcomes for patients.¹

However, P4P programs may have potential unintended consequences,^1,2,4 such as spillover effects on neglecting conditions or activities that are not directly rewarded.⁴ Holmstrom and Milgrom proposed the “multitasking problem” to illustrate the possibility of spillover effects.⁶ In a P4P scheme, healthcare providers might pay more attention to indicators tied to financial incentives and neglect those that are not incentivized. Therefore, P4P programs may improve the performance of incentivized items at the expense of the unfavorable performance of non-incentivized items, resulting in a decline in the quality of care for patients. This phenomenon is called a “negative spillover effect.”

The problem of multitasking might be alleviated if the activities undertaken in a process are similar or co-occur, because certain conditions or indicators rely on similar inputs.^7,8 This means that, under a P4P program, non-incentivized activities may be indirectly rewarded when they share commonalities with incentivized activities. This phenomenon is called a “positive spillover effect” (or halo effect). Therefore, commonalities in production among a set of activities are crucial for generating positive or negative spillover effects from an incentive program.

Although the intended effects of P4P programs have been well documented,^1-5 little is known about their spillover effects, and the results tend to be inconclusive. Some studies have found that P4P programs have a positive spillover effect on non-incentivized conditions or indicators,^9-12 whereas others have found negative spillover effects, because of the possible neglect of non-incentivized aspects of care by healthcare providers over the long term.^13,14 Some studies have found no spillover effect at all of P4P programs.^8,15 The aforementioned discrepancies in the findings might be attributed to variations in the types of P4P programs implemented, the quality of the methodology used, and differences in the research setting.

Most empirical studies on the spillover effects of P4P programs have been performed in the United States^8,11,15 or the United Kingdom.^9,10,13,14 The spillover effects of P4P programs in the United States have tended to be small, which might be attributable to the diluting effect of the small incentive size based on multiple payers and the phenomenon in which payers often consider only a fraction of the targeted healthcare providers.^2,3 The P4P programs in the United Kingdom have been implemented nationally. Most of these programs are designed for primary care settings and reward general practitioners based on the quality-of-service delivery. However, healthcare systems in many Asian countries such as Japan, Korea, and Taiwan lack general practitioners (as gatekeepers) or formal referral mechanisms and are mainly specialist-based care.

Methods

This study used a natural experimental design with population-based longitudinal data to examine the long-term spillover effects of a P4P program. The analysis was based on claims data for healthcare utilization from 2016 to 2020 in Taiwan.

Pay-for-Performance Program for Diabetes Care

Since the end of 2001, Taiwan’s single-payer, the National Health Insurance Administration (NHIA), has implemented nationwide P4P programs for several chronic conditions. For the P4P program for diabetes care, physicians who specialize in metabolic disorders or endocrinology or those who have completed training for the diabetes shared care program can voluntarily participate in the P4P program for diabetes care. Participating physicians can recruit patients for the program. The P4P program pays participating physicians three fees: the P4P management fee for the initial enrollment visit, an extra fee for a comprehensive follow-up visit, and an annual evaluation fee. The required and recommended services include a medical history examination, physical examination (for example, ophthalmoscopic or foot examination), laboratory evaluation (for example, HbA1c check or cholesterol checks), management plan, and diabetes self-management education, which are clearly defined in the P4P program for diabetes care.¹⁶

Outcome-based quality indicators have been gradually incorporated into the P4P program for diabetes care. Between 2001 and 2006, the financial incentives of the diabetes P4P program focused only on participation or process-based care (eg, performing HbA1c or cholesterol checks). In late 2006, the NHIA implemented a new strategy to better reward the health outcomes by paying an additional bonus for improvement in intermediate clinical outcome measures. A composite score was developed for each participating physician based on measures including the complete follow-up visit rate and control of HbA1c and low-density lipoprotein cholesterol (LDL-C) levels. Under this scheme, physicians receive an extra outcome-based reward (NT$ 1000 or US$ 33 per patient) if their composite scores rank among the top 25% of participating physicians. This plan is called the “pay for excellence” incentive. In 2009, the NHIA introduced a “pay-for-improvement” incentive to encourage physicians to focus on improving the composite scores of patients. Physicians receive an extra reward (NT$ 500 or US$ 17 per patient) if the composite score of a patient improves or is maintained over two years.¹⁶ In short, both process-based (eg, HbA1c check) and intermediate clinical outcome-based indicators (eg, proper control of HbA1c) for the recommended examination and laboratory tests have been incorporated into the diabetes P4P program. However, not all the intermediate clinical outcome indicators of the examinations/tests are incentivized in the program.¹⁶

Participants

Adult patients (aged 20 or older) diagnosed with type-2 diabetes were identified using the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) code E11. Because clinical data for intermediate outcomes (eg, HbA1c level) became available starting in 2015, we included only patients with at least three diabetes-related physician visits and laboratory test results every year from 2016 to 2020. Patients with diabetes who were enrolled in the nationwide P4P program in 2017 were included in the intervention group. The index date for each patient was defined as the date of enrollment in the P4P program between January 1, 2017 and December 31, 2017. The study included only participants who remained in the program throughout the observation period from 2017 to 2020.

Selection bias cannot be ruled out because the patients enrolled in the program were purposively selected by their physicians and voluntarily enrolled in the P4P program.^17,18 To minimize the potential impact of selection bias, we adopted a two-step sample selection: identifying patients visiting the same group of physicians and using the propensity score matching (PSM) approach to select patients. In the first step, we identified the most frequently visited physicians for the diabetes care of the 9119 patients enrolled in the P4P program. All diabetic patients who visited the shortlisted set of physicians were identified. Patients who had never been enrolled in the P4P program were considered suitable for the comparison group. A total of 2649 physicians were consulted by 9119 patients (pre-matched) in the intervention group, and 21 651 patients (pre-matched) were selected via PSM in the comparison group.

Subsequently, in the second step, we used a PSM approach to select appropriate patients to form the comparison group.¹⁹ For each patient, we created a propensity score that estimated the probability of their enrollment in the P4P program based on their characteristics using a generalized estimating equations (GEEs) model considering the effects of patient clustering among particular physicians.²⁰ The characteristics of patients considered in the matching process included their age, sex, Charlson comorbidity index (CCI) score,²¹ diabetes complication severity index (DCSI) score,²² likelihood of hospitalization in the previous year, and baseline HbA1c values. The characteristics of healthcare providers included in the model were the accreditation level²³ and location of the hospital or community clinic that was most frequently visited by a given patient.

Based on the propensity scores, we used the caliper matching method with 1:2 matching between the intervention and comparison groups. Finally, 7688 patients (post-matched) were enrolled from the P4P program, and the PSM process yielded 15 376 patients (post-matched) in the comparison group. In addition, we calculated absolute standardized differences in baseline characteristics between the intervention and comparison groups. Standardized differences less than 10% indicated acceptable matching.^24-26

The observation period ranged from one year before the index date in 2017 to three years of subsequent follow-up. As the subjects in the comparison group did not have an index date unlike those enrolled in the P4P program, they were assigned a pseudo-index date of their matched counterparts in the intervention group. In all, 23 064 patients and 92 256 patient years were included in the analysis. The unit of analysis was the number of patient-years.

Measurement of Variables

In terms of dependent variables, four intermediate clinical outcome indicators were included in this study: two incentivized and two non-incentivized indicators under the P4P scheme. In the analysis, the two incentivized indicators measured whether the patient had proper control of their HbA1c (HbA1c <7%) and LDL-C levels (LDL-C <100 mg/dL) during the study period. High-density lipoprotein cholesterol (HDL-C) and triglyceride (TG) levels are also crucial for the management of dyslipidemia in patients with diabetes.²⁷ Therefore, we included two non-incentivized indicators to measure whether the patient had proper control over their HDL-C (HDL-C >40 mg/dL) and TG levels (TGs <200 mg/dL).

The leading independent variables were patient enrollment in the P4P program, time dummy variables for the three years after the index date, and three interaction terms for the previously described variables. The following covariates were controlled in the regression models: patient characteristics (sex, age, CCI score, DCSI score, and likelihood of hospitalization in the previous year) and healthcare provider characteristics (accreditation level and location).

Statistical Analyses

To examine the spillover effects of the P4P program, we fitted the GEEs with difference-in-differences (DID) estimation to longitudinal data that considered the correlation between repeated observations for the same patients and patients within the same matched pairs as well as considered the unobserved time-invariant characteristics for patients.²⁰ The likelihood of the patient having proper control on the four intermediate clinical outcome indicators was analyzed using a logit link function and had a binominal distribution. The specifications are as follows:

logit Y_it = β₀ + β₁ × P4P_it + β₂ × Year1_it + β₃ × Year2_it + β₄ × Year3_it + β₅ × P4P_it × Year1_it + β₆ × P4P_it × Year2_it + β₇ × P4P_it × Year3_it + β₈ × X_it + β₉ × Z_i + ε_it

where Y_it is the likelihood of good control on an intermediate clinical outcome for patientiduring period t. The P4P_it dummy indicates whether the patient was enrolled in the P4P program. Year1_it, Year2_it, and Year3_it are dummy variables that indicate observations for over 3 years after the index date of the P4P program. The DID estimates of the effects of the P4P program are captured by the coefficients β₅, β₆, and β₇ on the three interaction terms P4P_it × Year1_it, P4P_it × Year2_it, and P4P_it × Year3_it, respectively. Parameters X_it and Z_i represent a set of variables that measure time-variant and time-invariant covariates (participant’s sex). All statistical analyses were performed using the SAS version 9.4 (SAS Institute) and Stata version 15.1 (StataCorp).

Results

Descriptive Analysis

Table 1 lists the baseline characteristics of the pre- and post-matched participants in the intervention and comparison groups. In the pre-matched sample, patients diagnosed with diabetes in the intervention group were younger, had higher DCSI and CCI scores, and controlled their baseline HbA1c levels less well. Furthermore, they had a higher likelihood of hospitalization in the previous year and tended to receive care at a medical center/regional hospital. The PSM process resulted in a more balanced distribution of the characteristics of patients and their care providers in the intervention and comparison groups. All absolute standardized differences were less than 10%, indicating acceptable matching results.

Figure presents the changes in the outcome measures over the study period. For the incentivized indicators, proper control of HbA1c levels was lower in the intervention group than in the comparison group in the baseline year (43.59% vs. 45.05%). After the first year of the P4P program, the rate of proper control of HbA1c increased in the intervention group (49.08%) and was higher than that in the comparison group (46.85%). For proper control of LDL-C, the rates in the intervention and comparison groups increased after the intervention (ranging from 56.18% to 70.51% in the intervention group; 57.30% to 66.77% in the comparison group). For non-incentivized indicators, we found that the rates of proper control on HDL-C steadily increased in both the intervention and comparison groups over four years. Finally, we found a noticeable increase in the rate of proper control of TG levels in the intervention group after enrollment, compared with a slight increase in the comparison group.

Close

Figure. Distribution of Intermediate Clinical Outcome Measures by Intervention Status Over Study Period. Abbreviations: P4P, pay-for-performance; HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglyceride.

Results From Regression Models

Table 2 lists the results obtained from the DID estimates in the GEEs model. For the incentivized indicator HbA1c, patients enrolled in the P4P program were more likely than their counterparts to have proper control of their HbA1c levels during the post-P4P period. The odd ratios (ORs) of the interaction terms were positive and significant from the first to third years after the intervention (OR was 1.166 in the first year [95% confidence interval [CI]: 1.106-1.230], 1.168 in the second year [95% CI: 1.103-1.236], and 1.138 in the third year [95% CI: 1.073-1.206]). In respect of the proper control of LDL-C, we also found that the DID parameters were positive and significant, with OR being 1.199 (95% CI: 1.128-1.273), 1.232 (95% CI: 1.155-1.314), and 1.231 (95% CI: 1.150-1.318) for the three years after the intervention.

For the non-incentivized indicator TGs, the result was similar to that of the aforementioned incentivized indicators. Patients in the P4P program were more likely than their counterparts to have proper control of their TG levels in the first and second years after the P4P program (OR of 1.114 [95% CI: 1.044-1.189], 1.122 [95% CI: 1.047-1.202]). However, there was no significant effect in the third year after the P4P program (OR of 1.051 [95% CI: 0.980-1.128]). Conversely, we found no spillover effects of the P4P program on the non-incentivized HDL-C control in the three years after the P4P program (OR of 1.036 in the first year [95% CI: 0.985-1.089]; 0.986 in the second year [95% CI: 0.936-1.040]; and 0.974 in the third year [95% CI: 0.922-1.029]) (Table 2).

Sensitivity Analysis

Sensitivity analyses were performed to validate the robustness of the findings. First, we used a continuous scale of the intermediate clinical outcome variables instead of a binary one. The analyses yielded results similar to those described earlier (Supplementary file 1, Table S1). Second, we performed analyses stratified by sex and found partly similar results (Table S2). Third, we performed analyses stratified by the status of multiple chronic conditions (<2 chronic conditions or ≥2 chronic conditions, excluding diabetes). In this study, the number of chronic conditions was measured using nine common chronic conditions that often require ongoing medication management, including hypertension, heart disease, cerebrovascular disease, pulmonary disease, chronic renal disease, arthritis/degenerative joint diseases, depression/anxiety-associated diseases, and cancer. Most of the results were similar to previous findings, except for patients with ≥2 chronic conditions, in terms of TG outcomes (Table S3). Finally, we used random-intercept models with DID estimation to consider unobserved time-invariant covariates and considered the correlation between repeated observations for the same patient. Instead of the odds ratios, we calculated the average marginal effects for all estimates and used bootstrapping with 100 replications to acquire standard errors (Table S4). Previously, a similar approach has been used in the econometric literature. These sensitivity analyses yielded results similar to our main results.

Discussion

Effect of P4P Programs on Incentivized Outcomes

The effect of a P4P program on intermediate clinical outcomes tended to be inconclusive in a previous systematic review.⁵ For example, under the Quality of Outcome Framework (QOF) in the United Kingdom, Vamos et al²⁸ and Alshamsan et al²⁹ performed an interrupted time series analysis and found that the trend in the proper control of HbA1c in patients with diabetes after the introduction of the QOF worsened relative to the pre-QOF trend. In contrast, we found that the P4P program for diabetes had a positive long-term impact on incentivized intermediate outcomes, including proper control of HbA1c and LDL-C levels in patients. Several possible explanations can explain this outcome. First, the universal QOF scheme is designed for primary care settings to reward general practitioners for achieving quality indicators. However, in Taiwan, P4P programs are designed with disease-specific incentives and rewards are given to specialists to achieve pre-determined targets. In other words, focusing on the management of diabetes care based on clinical guideline-recommended services (eg, medication management) could lead to better glycemic control. Second, the P4P scheme for diabetes in Taiwan provides multiple incentives. From 2001 to 2006, incentives were exclusively based on process-of-care outcomes (eg, receipt of HbA1c testing). Incentives targeting intermediate clinical outcomes (eg, proper control of HbA1c) were incorporated in 2006, and a bonus for improvement in intermediate clinical outcomes was introduced in 2009. Combining multiple incentive designs might better motivate the care providers toward improving intermediate clinical outcomes.¹

Spillover Effect of P4P Programs on Non-incentivized Outcomes

Previous studies have raised concerns about the unintended effects of P4P programs.^1,2,4 However, research on the spillover effects of P4P programs on non-incentivized clinical outcomes is limited, and the findings tend to be inconclusive.^8-15 This study focused on whether the intended improvement might spill over to other non-incentivized aspects of care, resulting in a decline in the overall quality of care among patients with diabetes. Under the P4P program for diabetes care in Taiwan, checks of patient TGs and HDL-C (a process of care) are incentivized, but control of patients’ TG and HDL-C levels (intermediate clinical outcomes) are not incentivized. For patients with diabetes, a high TG or low HDL-C level is associated with a higher risk of vascular complications.³⁰ Therefore, controlling patients’ TG and HDL-C levels is essential for diabetes management, which in turn affects the overall health outcome. The findings from this study suggest that the P4P did not affect the control of non-incentivized HDL-C levels among the intervention group even three years after enrollment. However, we found that the P4P program had a positive spillover effect on TG control (another non-incentivized parameter) in the first and second years after the P4P program.

The inconsistent findings of the spillover effects seem intriguing; however, when we consider “commonality in production,” these seem reasonable. In the P4P program, the multitasking problem can be mitigated when the procedures used by physicians to address the incentivized and non-incentivized indicators are similar.^7,8 We observed that the improvement in TG control (a non-incentivized outcome) occurred because it was addressed by physicians in a manner similar to that of HbA1c control (an incentivized outcome). The recommendations of the American Diabetes Association emphasize that lifestyle interventions and glycemic control are both beneficial for improving TG control.²⁷ For example, metformin is a first-line pharmacological agent used for glycemic control in type-2 diabetes, and evidence has shown that metformin can significantly decrease TG levels in patients with diabetes.³¹ In other words, for patients enrolled in the P4P program for diabetes care, physicians might regularly monitor their blood glucose levels and improve their glycemic control using medications such as metformin, which also improves the TG level. However, the improvement in HDL-C control was not affected by the use of similar treatments. The effects of metformin on HDL-C levels are less evident.^32,33 Lifestyle interventions such as weight loss, exercise, and diet control are better ways to manage HDL-C levels.²⁷ Therefore, our findings support the idea that commonalities among a set of activities are crucial for financial incentive programs to generate positive or no spillover effects.^8,12

Limitations

This study had some limitations. First, the participants were not randomly assigned to the intervention or comparison groups, which may have caused selection bias. In this study, we used two-step sample selection strategies to minimize selection bias: selecting patients from the same physician and using PSM. Although the PSM matching strategy may have increased the similarity of participants in the intervention and comparison groups at baseline, it does not ensure that the two groups were similar before the intervention. Owing to the lack of available intermediate clinical outcome data, we performed a placebo DID analysis using the participants’ diabetes-related hospitalizations as proxies to examine the parallel trend assumption before the intervention. During 2007 and 2016 (before the P4P intervention in 2017), the placebo DID demonstrated small and insignificant changes in hospitalization for diabetes-related conditions, which implied similar trends in the two groups before the intervention (Table S5).³⁴ Finally, the participants in this study were representative of patients with at least three diabetes-related physician visits and laboratory test results, which might not represent all diabetes patients in Taiwan.

Conclusion

This is one of the first studies to evaluate the long-term spillover effects of a P4P program on intermediate clinical outcomes under a single-payer healthcare system. This study provides evidence showing that the P4P program has significantly improved incentivized outcome measures. In addition, the P4P program has a positive spillover effect on a non-incentivized outcomes when the related measure is treated in a manner similar to that of the incentivized indicators. However, this study found no spillover effect on non-incentivized outcomes when the related measure is not treated in a manner similar to that of incentivized indicators. We speculate that the problem of multitasking can be mitigated if a set of indicators is addressed using similar treatment measures. Therefore, we suggest that the incentive design of P4P programs should consider “commonality in procedure” to facilitate positive spillover effects and avoid negative spillover effects. Future investigations are necessary to examine the spillover effects of a financial incentive program with both “targeted vs. non-targeted conditions” and “incentivized vs. non-incentivized indicators” under various health-care settings and adopt a more extended follow-up.

Ethical issues

This study was approved by the Institutional Review Board of Fu Jen Catholic University (C106074).

Competing interests

Authors declare that they have no competing interests.

Funding

The study was supported by grants from the National Science and Technology Council (NSTC 112-2410-H-002 -039 -MY3) in Taiwan. The funding source had no role in the study.

Supplementary files

References

1. Petersen LA, Woodard LD, Urech T, Daw C, Sookanan S. Does pay-for-performance improve the quality of health care?. Ann Intern Med 2006; 145(4):265-272. doi: 10.7326/0003-4819-145-4-200608150-00006 [Crossref] [ Google Scholar]
2. Rosenthal MB, Frank RG. What is the empirical basis for paying for quality in health care?. Med Care Res Rev 2006; 63(2):135-157. doi: 10.1177/1077558705285291 [Crossref] [ Google Scholar]
3. Van Herck P, De Smedt D, Annemans L, Remmen R, Rosenthal MB, Sermeus W. Systematic review: effects, design choices, and context of pay-for-performance in health care. BMC Health Serv Res 2010; 10:247. doi: 10.1186/1472-6963-10-247 [Crossref] [ Google Scholar]
4. Eijkenaar F, Emmert M, Scheppach M, Schöffski O. Effects of pay for performance in health care: a systematic review of systematic reviews. Health Policy 2013; 110(2-3):115-130. doi: 10.1016/j.healthpol.2013.01.008 [Crossref] [ Google Scholar]
5. Mendelson A, Kondo K, Damberg C. The effects of pay-for-performance programs on health, health care use, and processes of care: a systematic review. Ann Intern Med 2017; 166(5):341-353. doi: 10.7326/m16-1881 [Crossref] [ Google Scholar]
6. Holmstrom B, Milgrom P. Multitask principal-agent analyses: incentive contracts, asset ownership, and job design. J Law Econ Organ 1991; 7:24-52. doi: 10.1093/jleo/7.special_issue.24 [Crossref] [ Google Scholar]
7. Glazer J, McGuire T, Normand SL. Mitigating the problem of unmeasured outcomes in quality reports. B E J Econom Anal Policy 2008; 8(2):7. doi: 10.2202/1935-1682.1738 [Crossref] [ Google Scholar]
8. Mullen KJ, Frank RG, Rosenthal MB. Can you get what you pay for? Pay-for-performance and the quality of healthcare providers. Rand J Econ 2010; 41(1):64-91. doi: 10.1111/j.1756-2171.2009.00090.x [Crossref] [ Google Scholar]
9. Campbell S, Reeves D, Kontopantelis E, Middleton E, Sibbald B, Roland M. Quality of primary care in England with the introduction of pay for performance. N Engl J Med 2007; 357(2):181-190. doi: 10.1056/NEJMsr065990 [Crossref] [ Google Scholar]
10. Steel N, Maisey S, Clark A, Fleetcroft R, Howe A. Quality of clinical primary care and targeted incentive payments: an observational study. Br J Gen Pract 2007; 57(539):449-454. [ Google Scholar]
11. Sutton M, Elder R, Guthrie B, Watt G. Record rewards: the effects of targeted quality incentives on the recording of risk factors by primary care providers. Health Econ 2010; 19(1):1-13. doi: 10.1002/hec.1440 [Crossref] [ Google Scholar]
12. Sherry TB, Bauhoff S, Mohanan M. Multitasking and heterogeneous treatment effects in pay-for-performance in health care: evidence from Rwanda. Am J Health Econ 2017; 3(2):192-226. doi: 10.1162/AJHE_a_00072 [Crossref] [ Google Scholar]
13. Campbell SM, Reeves D, Kontopantelis E, Sibbald B, Roland M. Effects of pay for performance on the quality of primary care in England. N Engl J Med 2009; 361(4):368-378. doi: 10.1056/NEJMsa0807651 [Crossref] [ Google Scholar]
14. Doran T, Kontopantelis E, Valderas JM. Effect of financial incentives on incentivised and non-incentivised clinical activities: longitudinal analysis of data from the UK Quality and Outcomes Framework. BMJ 2011; 342:d3590. doi: 10.1136/bmj.d3590 [Crossref] [ Google Scholar]
15. Glickman SW, Ou FS, DeLong ER. Pay for performance, quality of care, and outcomes in acute myocardial infarction. JAMA 2007; 297(21):2373-2380. doi: 10.1001/jama.297.21.2373 [Crossref] [ Google Scholar]
16. Cheng SH, Lin YY. Pay-for-performance programs in Taiwan. In: Hu TW eds. Health Care Policy in East Asia: A World Scientific Reference. Volume 4: Health Care System Reform and Policy Research in Taiwan. New Jersey: World Scientific; 2020:157-180. 10.1142/10794-vol4.
17. Shen Y. Selection incentives in a performance-based contracting system. Health Serv Res 2003; 38(2):535-552. doi: 10.1111/1475-6773.00132 [Crossref] [ Google Scholar]
18. Chang RE, Lin SP, Aron DC. A pay-for-performance program in Taiwan improved care for some diabetes patients, but doctors may have excluded sicker ones. Health Aff (Millwood) 2012; 31(1):93-102. doi: 10.1377/hlthaff.2010.0402 [Crossref] [ Google Scholar]
19. Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983; 70(1):41-55. doi: 10.1093/biomet/70.1.41 [Crossref] [ Google Scholar]
20. Fitzmaurice GM, Laird NM, Ware JH. Applied Longitudinal Analysis. Hoboken, NJ: John Wiley & Sons; 2004.
21. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992; 45(6):613-619. doi: 10.1016/0895-4356(92)90133-8 [Crossref] [ Google Scholar]
22. Young BA, Lin E, Von Korff M. Diabetes complications severity index and risk of mortality, hospitalization, and healthcare utilization. Am J Manag Care 2008; 14(1):15-23. [ Google Scholar]
23. Huang P, Hsu YH, Kai-Yuan T, Hsueh YS. Can European external peer review techniques be introduced and adopted into Taiwan’s hospital accreditation system?. Int J Qual Health Care 2000; 12(3):251-254. doi: 10.1093/intqhc/12.3.251 [Crossref] [ Google Scholar]
24. Normand ST, Landrum MB, Guadagnoli E. Validating recommendations for coronary angiography following acute myocardial infarction in the elderly: a matched analysis using propensity scores. J Clin Epidemiol 2001; 54(4):387-398. doi: 10.1016/s0895-4356(00)00321-8 [Crossref] [ Google Scholar]
25. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med 2009; 28(25):3083-3107. doi: 10.1002/sim.3697 [Crossref] [ Google Scholar]
26. Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res 2011; 46(3):399-424. doi: 10.1080/00273171.2011.568786 [Crossref] [ Google Scholar]
27. Haffner SM. Dyslipidemia management in adults with diabetes. Diabetes Care 2004; 27 Suppl 1:S68-71. doi: 10.2337/diacare.27.2007.s68 [Crossref] [ Google Scholar]
28. Vamos EP, Pape UJ, Bottle A. Association of practice size and pay-for-performance incentives with the quality of diabetes management in primary care. CMAJ 2011; 183(12):E809-816. doi: 10.1503/cmaj.101187 [Crossref] [ Google Scholar]
29. Alshamsan R, Lee JT, Majeed A, Netuveli G, Millett C. Effect of a UK pay-for-performance program on ethnic disparities in diabetes outcomes: interrupted time series analysis. Ann Fam Med 2012; 10(3):228-234. doi: 10.1370/afm.1335 [Crossref] [ Google Scholar]
30. Wong NKP, Nicholls SJ, Tan JTM, Bursill CA. The role of high-density lipoproteins in diabetes and its vascular complications. Int J Mol Sci 2018; 19(6):1680. doi: 10.3390/ijms19061680 [Crossref] [ Google Scholar]
31. Wulffelé MG, Kooy A, de Zeeuw D, Stehouwer CD, Gansevoort RT. The effect of metformin on blood pressure, plasma cholesterol and triglycerides in type 2 diabetes mellitus: a systematic review. J Intern Med 2004; 256(1):1-14. doi: 10.1111/j.1365-2796.2004.01328.x [Crossref] [ Google Scholar]
32. Bolen S, Feldman L, Vassy J. Systematic review: comparative effectiveness and safety of oral medications for type 2 diabetes mellitus. Ann Intern Med 2007; 147(6):386-399. doi: 10.7326/0003-4819-147-6-200709180-00178 [Crossref] [ Google Scholar]
33. Chaudhuri A, Dandona P. Effects of insulin and other antihyperglycaemic agents on lipid profiles of patients with diabetes. Diabetes ObesMetab 2011; 13(10):869-879. doi: 10.1111/j.1463-1326.2011.01423.x [Crossref] [ Google Scholar]
34. Huntington-Klein N. The Effect: An Introduction to Research Design and Causality. 1st ed. Boca Raton, FL: CRC Press; 2022.