Early Enrollment in Diabetes Pay-for-Performance Program Reduced Loss of Life Expectancy in Newly-Diagnosed Patients with Type 2 Diabetes Mellitus

Article information

Diabetes Metab J. 2025;.dmj.2024.0507

Publication date (electronic) : 2025 March 26

doi : https://doi.org/10.4093/dmj.2024.0507

Yu-Ching Chen^1,2, Wei-Ming Wang³, Boniface J. Lin^4,5, Jung-Der Wang^,1,6, Li-Jung Elizabeth Ku^,1

¹Department of Public Health, College of Medicine, National Cheng Kung University, Tainan, Taiwan

²Department of Healthcare Administration and Medical Informatics, Kaohsiung Medical University, Kaohsiung, Taiwan

³Department of Statistics and Information Science, Fu Jen Catholic University, New Taipei City, Taiwan

⁴College of Medicine, Fu Jen Catholic University, New Taipei, Taiwan

⁵College of Medicine, National Taiwan University, Taipei, Taiwan

⁶Department of Occupational and Environmental Medicine, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan

Corresponding authors: Li-Jung Elizabeth Ku https://orcid.org/0000-0002-4644-8439 Department of Public Health, College of Medicine, National Cheng Kung University, No. 1, University Road, Tainan 701, Taiwan E-mail: eljku@mail.ncku.edu.tw

Jung-Der Wang https://orcid.org/0000-0002-3176-4500 Department of Public Health, College of Medicine, National Cheng Kung University, No. 1, University Road, Tainan 701, Taiwan E-mail: jdwang121@gmail.com

Received 2024 August 25; Accepted 2024 December 3.

Abstract

GRAPHICAL ABSTRACT

Highlights

INTRODUCTION

Diabetes is a disease of increasing prevalence with 537 million worldwide and 90 million in South-East Asia in 2021 [1]. Years spent with diabetes increased by 156% in men and 70% in women in the USA during 1985–2011 [2]. Type 2 diabetes mellitus was associated with an average 6-year loss of life expectancy (LE) [3], more among whites than among South Asians [4].

Early diagnosis and continuing multidisciplinary care with regular monitoring and timely intervention are the key elements in reducing the disease burden of diabetes [5]. Intensive glucose control starting at the time of diagnosis was associated with a significantly decreased risk of death from any cause [6], while intensive glucose lowering in patients with 10-year median duration of diabetes might increase mortality [7]. It appeared that time interval between diagnosis and target-driven treatment could predict survival.

In August, 1996, a pilot study of a modified “Diabetes Shared Care Network” was launched at Yilan county, Taiwan, and later expanded to three other counties; in 2001, Taiwan implemented a pay-for-performance (P4P) program islandwide (Guaynabo, Puerto Rico) [8], providing multidisciplinary team care to prevent diabetes complications. In a systematic review of the outcome effect of P4P program on diabetes in single-payer national health systems [9], it was associated with a lower risk of mortality in some studies of Taiwan [8,10,11], in one study [12] but not in the other study [13] of England or in one study [14] of Canada. In South Korea, the achievement of the national diabetes quality assessment program indicators was associated with a decreased risk of all-cause mortality [15].

Rarely has any study been done to investigate the impact of P4P program on LE in patients with diabetes. To control potential confounding by different age, gender, and calendar year of care technology for diabetes, we employed age, gender, and calendar year-matched referents simulated from vital statistics of general population to estimate the loss-of-LE as the indicator to determine if early enrollment in P4P program could prolong the survival in patients with diabetes. Namely, we aimed to (1) compare years of life lost associated with the interval between diabetes diagnosis and enrollment in P4P program and (2) quantify these relationships in subgroups of different intervals.

METHODS

Data source and study cohort

Data source

Taiwan introduced a compulsory, single-payer National Health Insurance (NHI) program with universal coverage in 1995. By 2009, approximately 99% of Taiwan’s residents had been enrolled in the NHI program [16]. From the original NHI database, the Ministry of Health and Welfare has provided two main cohort categories: population-based and disease-specific. The latter aimed to facilitate the rapid identification of patients with specific disease. We used the disease-specific diabetes mellitus health database, where patients with diabetes, by ICD-9-CM code 250 or ICD-10-CM code E11, were newly diagnosed based on at least one hospital admission or three or more outpatient visits within a calendar year from 2004 to 2018 [17]. They were interlinked with two other databases, the Taiwan Mortality Registry (2004–2018) and the Taiwan National Vital Statistics (2004–2018), as presented in Fig. 1. The study was conducted after obtaining approval by the Human Research Ethics Committee of National Cheng Kung University (HREC [Exempt] No. 111-493), and the requirement for informed consent was waived.

Fig. 1.

Flow diagram of establishment of subcohorts and estimation for life expectancy (LE) and loss-of-LE. T2DM, type 2 diabetes mellitus; ICD-9 or 10-CM, International Classification of Diseases, 9th or 10th Revision, Clinical Modification; P4P, pay-forperformance. aInterval between diabetes diagnosis and enrollment in P4P program.

Study cohort

Fig. 1 illustrates the process of establishing the study cohort. Patients with type 1 diabetes mellitus were excluded from this study. Derived from the sum of newly-diagnosed patients with diabetes, our study population began with 2,040,597 patients with diabetes during 2001–2015. The exclusion criteria were as follows: (1) Diabetes was diagnosed before the end of 2003 (n=315,359); (2) those with missing data in gender (n= 59,164); (3) those with age <40 or ≥65 years (n=651,348); (4) those with death date before diabetes diagnosis (n=3); (5) those with any catastrophic illness before diabetes diagnosis (n=60,990). Among the remaining 953,733 patients, 503,662 were enrolled in the P4P program (internal code in the NHI system: P14xx) between January 1, 2004 and December 31, 2015. As we recruited patients with diabetes into this cohort by calendar year during 2004–2015 and followed them until 2018, patients diagnosed in January 2004 could be followed for 15 years, whereas those diagnosed in December 2015 might be followed for 3 years only.

From mortality registry to loss-of-LE (‘Statistical analysis’ section), we further stratified the P4P group into four subgroups according to the time interval between diabetes diagnosis and enrollment in the P4P program: interval <1, 1–4, 5–9, and 10–15 years. The interval was calculated for an individual from the date of newly-diagnosed diabetes to the date of enrollment in P4P program confirmed by the first appearance of code P1407C in the claim data.

Availability of data and materials

The data that support the findings of this study are available from the Health and Welfare Data Science Center of Taiwan, but restrictions apply to the availability of these data, which were applied to be used exclusively for the current study, and so are not publicly available.

Statistical analysis

Estimation of life expectancy, loss of life expectancy, and life-year gained

By interlinking the Taiwan National Mortality Registry and the reimbursement database of NHI for 503,662 subjects in P4P group and 450,071 subjects in non-P4P group, we applied the Kaplan–Meier method to estimate the survival function in the follow-up period of 15 years, namely 2004–2018, and then extrapolated to lifetime by a semiparametric method [21]. Namely, we applied Monte Carlo methods to obtain the lifetime survival function from age-, gender- and calendar yearmatched referents simulated from the National Vital Statistics database. With premature mortality caused by diabetes, the relative survival between each group and its corresponding reference population ranged between 0 and 1. Next, we performed logit transformations of the relative survivals obtained from the above methods and constructed a restricted cubic splines model to assure linear beyond the last knot. We used the fitted model to extrapolate the first month beyond 15-year follow-up, which was generally accurate and was treated as a new ‘observed value.’ Month-by-month, we repeatedly updated the data of the logit-transformed function by dropping the first month and adding the new ‘observed value’ as the last value, and re-fitted the restricted cubic splines model to obtain updated model for predicting a new value in the immediate next month. Through the above month-by-month rolling-over algorithm, we extrapolated the survival of the index group down until below 1% to stop (Fig. 2). Thereby, we estimated the lifetime survival, and the area under the survival curve was the LE of the patients in each index subcohort. We subtracted the LE of each index cohort from the corresponding age-, gender- and calendar year-matched reference population to obtain the loss-of-LE. The standard errors of LE and loss-of-LE were generated through a bootstrap method by implementing the extrapolation process with data simulated by repeatedly sampling with replacement from the real data set 100 times. The rolling extrapolation algorithm of iSQoL2 R package (http://sites.stat.sinica.edu.tw/isqol) used in this study is based on the estimated survival function derived from the observed survival times of all patients with diabetes in non-P4P group, P4P group and four subgroups of P4P group, respectively, during the followup period [21]. The survival times of patients with diabetes who were still alive at the end of the follow-up period, namely the end of 2018, were treated as censored, regardless of various lengths of their follow-up. Specifically, the Kaplan–Meier method was employed to estimate the survival function from the survival data. A key advantage of the Kaplan–Meier curve is its ability to incorporate these censored data effectively.

Fig. 2.

This figure shows estimation of lifetime survival function in non-pay-for-performance (P4P) group (A), P4P group (B), and its four subgroups (C, D, E, F), compared with its corresponding reference population, respectively. We subtracted the life expectancy (LE) of each index cohort from that of its corresponding age-, gender- and calendar year-matched reference population to obtain the loss-of-LE. LE and loss-of-LE of each index cohort were shown numerically in its corresponding panel.

In addition, we used this longitudinal cohort data to determine the incidence of adverse outcomes, including myocardial infarction, stroke, and end-stage renal disease (ESRD) (‘Comorbidities’ part of ‘Measures’ section), and catastrophic illnesses, DCSI and insulin use (‘Other covariates’ part of ‘Measures’ section) until death or the end of 2018, which was presented in Tables 1 and 2. We performed descriptive analyses where the continuous variables were presented as means and standard deviations, and dichotomized variables as percentages. The differences between the two compared groups were tested by t-test for continuous variables and chi-square test for categorical variables. Test for a linear trend was also performed for stratifications of age strata and for strata of intervals preceding enrollment in P4P program after diabetes diagnosis. Descriptive statistical analyses and survival analyses were performed using the SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA). The above rolling extrapolations to estimate LEs and loss-of-LEs were carried out by the iSQoL2 R package developed by Hwang and Wang [22].

Table 1.

Life expectancy and loss of life expectancy in patients with diabetes with catastrophic illness, comorbidities, and insulin use

Table 2.

Stratification by P4P status, interval between diabetes diagnosis and enrollment in P4P program and diabetes complications severity index

RESULTS

Table 3 compares the characteristics of the study cohort before diabetes diagnosis, including 503,662 subjects in P4P group and 450,071 subjects in non-P4P group; P4P group was stratified into four subgroups based on interval: 164,711 for interval less than 1 year, 163,313 for interval 1–4 years, 134,620 for interval 5–9 years, and 41,018 for interval 10–15 years, respectively. Mean age was 53.0 to 54.0 years, and 43.3%–45.7% were women across non-P4P group and P4P subgroups. Significant P value occurred among four subgroups of P4P group across comorbidities, except depression.

Table 3.

Distribution of study cohort by gender, age, Charlson comorbidity index, and comorbidities

In Table 1, when P4P group was stratified by interval between diabetes diagnosis and enrollment in P4P program, the result showed that loss-of-LE was 0.06±0.72 years for interval less 1 year, 0.05±0.59 years for interval 1–4 years, 10.01±0.11 years for interval 5–9 years, and 12.77±0.14 years for interval 10–15 years, respectively. When the difference in loss was measured between with and without comorbidity, namely catastrophic illness, stroke, ESRD, or myocardial infarction. There was significant difference (P<0.001) among the four subgroups of P4P group where the difference in loss was large in intervals less than 1 and 1–4 years, and became smaller and smaller from interval 5–9 years to interval 10–15 years, suggesting that someone did not have one comorbidity, while the other comorbidity or more could have developed, leading to large loss-of-LE. In terms of insulin use, subgroup of interval 5–9 years suffered greater loss-of-LE, namely, 2.34±1.96 vs. 0.02±0.48 years without insulin and 13.21±0.20 vs. 7.61±0.30 years with insulin, compared with subgroups of interval <5 years.

Table 2 summarizes the loss-of-LE in each subcohort after stratification by scores of DCSI. We found that P4P group enrolled within 5 years after diabetes diagnosis did show higher proportions of DCSI 1–2 (18.4% vs. 14.6%) with nearly no loss-of-LE, but lower proportion of DCSI ≥3 (67.4% vs. 76.5%) with less loss-of-LE (2.33 years vs. 10.43 years), compared with those enrolled later. Furthermore, those enrolled after 5 years of diabetes diagnosis showed longer diabetes duration accompanied with shorter P4P care. Moreover, a large loss-of-LE was associated with DCSI ranging from 0 to 2 in those patients.

Fig. 2 shows estimation of lifetime survival function in non-P4P group, P4P group and its four subgroups, compared with its corresponding reference population. Loss-of-LE was 2.60 years for non-P4P group, 3.55 years for all P4P group, 0.06 year for interval less than 1 year, 0.05 year for interval 1–4 years, 10.01 years for interval 5–9 years, and 12.77 years for interval 10–15 years, respectively.

Fig. 3 demonstrates the validation of survival function by comparing the result of extrapolation from the end of the 8th year to the 15th year with that of the actual survival rate based on Kaplan–Meier’s estimate throughout 15 years of follow-up. The relative bias assuming the Kaplan–Meier’s estimate as the gold standard was 2.5%.

Fig. 3.

This figure demonstrates the validation of survival function by comparing the result of extrapolation from the end of the 8th year to the end of the 15th year with that of the actual survival rate based on Kaplan–Meier’s (K-M) estimate throughout 15 years of follow-up. The relative bias assuming the K-M estimate as the gold standard was 2.5%.

In Supplementary Table 4, female showed a higher percentage (44.7% vs. 43.3%) in P4P group than in non-P4P group. The mean age of diabetes diagnosis was 53.6 years for P4P group and 53.7 years for non-P4P group, respectively. For CCI score, P4P group vs. non-P4P group is 53.2% vs. 51.9% for score 0, 31.3% vs. 31.5% for score 1, and 10.8% vs. 11.3% for score 2. Significant difference occurred across comorbidities except coronary artery disease, rheumatology disorders and psychosis.

In Supplementary Table 5, during 15-year follow-up, mean duration of diabetes was 8.16 years for non-P4P group and 9.16 years for P4P group; mean duration of P4P status was 5.34 years. P4P group (n=503,662) was associated with significantly more loss-of-LE (3.55±0.25 years vs. 2.60±0.14 years, P<0.001), compared with non-P4P group (n=450,071). In terms of difference in loss-of-LE, P4P group was associated with less difference in loss-of-LE in patients with catastrophic illness (P<0.001), but more difference in loss-of-LE in patients with myocardial infarction (P<0.001).

DISCUSSION

We found loss-of-LE was 0.06 year for interval less than 1 year, 0.05 year for interval 1–4 years, 10.01 years for interval 5–9 years, and 12.77 years for interval 10–15 years, respectively (P<0.001), compared with 2.60 years for the non-P4P group. Although our study seemed to provide evidence for the long-term or lifetime survival benefit for P4P program in diabetes care, our results of the survival benefit from P4P program could not be directly considered as causal. We have following arguments to corroborate the association between early enrollment in P4P program and reduced loss-of-LE. First, we used islandwide diabetes mellitus health database. This made our big study cohort highly representative of Taiwan’s whole diabetes population. The linkage with Taiwan Mortality Registry assured the comprehensiveness of their survival states. Second, high risk of all-cause mortality was associated with early age at diagnosis of diabetes [3] and intensive glucose lowering in patients with 10-year median duration of diabetes [7], while 10 years after the cessation of randomized intervention, a legacy effect was associated with intensive glucose control in patients with newly-diagnosed type 2 diabetes mellitus [6]. As diabetic complications increased with diabetes duration [23], elevated HbA1c [24], high blood pressure [25] and dyslipidemia [26], our findings (Table 1) on the probable impact of early enrollment in P4P program by stratification analysis corroborated these previous reports. Taiwan’s P4P program was target-driven multidisciplinary team care to empower patients for lifestyle modification [27], which corroborated the benefits of multifactorial risk reduction and improved survival as demonstrated in the Steno-2 study [28] with a median gain of 7.9 years in life [29]. Third, P4P subgroup of interval less than 1 year seemed to have a lower prevalence of comorbidities at baseline, suggesting an unintended consequence of P4P program, namely the exclusion of high-risk patients. Hsieh et al. [30] reported that patients who were male, older than 75 years, and those with DCSI scores or CCI scores larger than 2 were more likely to be excluded from enrollment, which could be gradually mitigated by increasing health literacy and growing enrollment rates. Therefore, early enrollment in Taiwan’s P4P program could achieve strict adherence, regular monitoring and timely intervention, and result in reduced loss-of-LE in the P4P group. Fourth, since we controlled for the potential confounding factors of gender, age, interval, major comorbidities (through catastrophic illness, stroke, ESRD, myocardial infarction, and DCSI), and insulin use in stratification analysis, the above factors could not explain the estimated difference in loss-of-LE between non-P4P and P4P subgroups (Table 1). Furthermore, in terms of comorbidities at baseline, non-P4P group and subgroup of interval 1–4 years had similar distributions of prevalence rates; among those involving target organ damages, namely coronary artery disease, heart failure, stroke, peripheral vascular disease and renal disease, the differences in prevalence across non-P4P group and four subgroups of P4P group were all less than or equal to 1.2% (Table 3), which corroborated the hypothesis of beneficial impact of early enrollment on LE. Fifth, in Table 1, the subgroup of interval 5–9 years had higher rates of catastrophic illnesses (17% vs.14%), stroke (10% vs. 6%), ESRD (3% vs. 2%), and myocardial infarction (4% vs. 3%), compared with subgroup of interval 1–4 years, and, to a greater degree, catastrophic illnesses (17% vs.11%), stroke (10% vs. 5%), ESRD (3% vs. 1%), and myocardial infarction (4% vs. 2%), compared with non-P4P group. Furthermore, the diabetes duration of subgroup of interval 5–9 years was 2 years longer than those of subgroup of interval 1–4 years and non-P4P group: 10.79±2.44 years vs. 8.05±3.08 and 8.16±3.50 years. Similarly, the difference in loss-of-LE among patients receiving insulin versus no insulin in subgroup of interval 5–9 years was higher than those of subgroup of interval 1–4 years and non-P4P group (10.87±1.98 years vs. 7.59±0.54 and 10.77±0.22 years). All these facts implied that early enrollment in the P4P program was associated with shorter duration of diabetes, timely insulin use [31,32], and fewer complications, corroborating Yoo et al.’s [33] observations. Sixth, as the validation of our extrapolation was demonstrated by an algorithm of repeatedly fitting restricted cubic spline models and rolling-over month-by-month from the end of the 8th year to the 15th year (Fig. 3) [34,35], we anticipated that the same method using 15 years of real data extrapolated to lifetime would be reasonably accurate. In fact, our approach of external additive hazards using the general population mortality as a splint for lifetime extrapolation was considered as relatively accurate in a recent review of statistical methods by van Oostrum et al. [36]. Thus, we tentatively concluded that the early enrollment (namely <5 years) in Taiwan’s P4P program of diabetes care was associated with a significant reduced loss-of-LE, which deserved further corroboration.

P4P program benefited the patients with diabetes with disability or depression. Among the disabled patients with diabetes, the risk of death was 0.41 times lower in those enrolled in the P4P program than in those who were not, suggesting that enrollment in the P4P program was beneficial in mitigating the risk of death [37]. Moreover, patients with diabetes with depressive symptoms enrolled in the diabetes P4P program had lower expenses for depression-related care than those not enrolled in the program, indicating a positive spillover effect of P4P program [38].

Lastly and importantly, stratification analysis of P4P subgroups revealed that loss-of-LE for subgroups of interval 5–9 and 10–15 years were 10.0 and 12.8 years, respectively, in contrast to almost no loss-of-LE in those who were enrolled in P4P within less than 5 years after diagnosis (Table 1). It corroborated that early diagnosis of diabetes and prompt enrollment in P4P program were essential to robustly implement target-driven multidisciplinary approach to ensure regular monitoring and timely intervention to reduce disease burden and prolong life in patients with diabetes.

There are some limitations in this study that must be acknowledged. First, this is not a randomized control trial; therefore, some unmeasured confounders cannot be ruled out despite stratification analysis. Second, limiting P4P enrollees to those aged 40 to 64 years, this study cannot be generalized to all patients with diabetes. Third, as there were no detailed data of HbA1c, blood pressure, LDL level, body mass index, cigarette smoking, physical activity, and educational level in claim data of NHI database, we were unable to make any additional pathophysiological inference. Future studies are warranted to dig into detailed mechanisms.

In conclusion, early enrollment in the P4P program was associated with reduced loss-of-LE, indicating P4P program might help patients with diabetes gain life if implemented early and faithfully after diabetes diagnosis.

SUPPLEMENTARY MATERIALS

Article information Continued