Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis

G Model

ARTICLE IN PRESS

HEAP-3816; No. of Pages 8

Health Policy xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Health Policy journal homepage: www.elsevier.com/locate/healthpol

Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis M. Ruth Lavergne a,∗ , Michael R. Law b , Sandra Peterson b , Scott Garrison c , Jeremiah Hurley d , Lucy Cheng b , Kimberlyn McGrail b a

Faculty of Health Sciences, Simon Fraser University, Blusson Hall, Room 10502, 8888 University Drive, Burnaby, BC V5A 1S6, Canada Centre for Health Services and Policy Research, School of Population and Public Health, Faculty of Medicine, University of British Columbia, 2206 E Mall, Vancouver, BC V6T 1Z3, Canada c Department of Family Medicine, University of Alberta, 6-60 University Terrace, Edmonton, AB T6G 2T4, Canada d Department of Economics, and Centre for Health Economics and Policy Analysis, McMaster University, 1280 Main Street West, Hamilton, Ontario, L8S 4K1, Canada b

a r t i c l e

i n f o

Article history: Received 29 November 2016 Received in revised form 11 October 2017 Accepted 2 November 2017 Keywords: Incentives in health care Chronic disease Primary care Administrative data uses Time series analysis

a b s t r a c t We studied the effects of incentive payments to primary care physicians for the care of patients with diabetes, hypertension, and Chronic Obstructive Pulmonary Disease (COPD) in British Columbia, Canada. We used linked administrative health data to examine monthly primary care visits, continuity of care, laboratory testing, pharmaceutical dispensing, hospitalizations, and total h ealth care spending. We examined periods two years before and two years after each incentive was introduced, and used segmented regression to assess whether there were changes in level or trend of outcome measures across all eligible patients following incentive introduction, relative to pre-intervention periods. We observed no increases in primary care visits or continuity of care after incentives were introduced. Rates of ACR testing and antihypertensive dispensing increased among patients with hypertension, but none of the other modest increases in laboratory testing or prescriptions dispensed reached statistical significance. Rates of hospitalizations for stroke and heart failure among patients with hypertension fell relative to pre-intervention patterns, while hospitalizations for COPD increased. Total hospitalizations and hospitalizations via the emergency department did not change. Health care spending increased for patients with hypertension. This large-scale incentive scheme for primary care physicians showed some positive effects for patients with hypertension, but we observe no similar changes in patient management, reductions in hospitalizations, or changes in spending for patients with diabetes and COPD. © 2017 Published by Elsevier Ireland Ltd.

1. Introduction Incentive payments aimed at improving healthcare delivery have been widely implemented, and many target chronic disease management in primary care [1]. Evidence of the impact of incentive programs on processes of care [2–11] and health outcomes [3,4,7,12] is mixed, though effects, where observed, are typically modest. For some interventions it is difficult to disentangle the effect of new incentive payments from other contemporaneous changes to the delivery of primary care, such as new teambased models and enhanced care coordination, or other quality

∗ Corresponding author. E-mail address: ruth [email protected] (M.R. Lavergne).

improvement efforts, notably performance measurement and public reporting [2,8,10]. Despite a large body of research, not all chronic conditions are well represented. For example, the impact of financial interventions on chronic obstructive pulmonary disease (COPD) management has not been examined [2–9,13]. Systematic reviews have concluded that more research on the impact of financial incentives is still needed [2–9]. The Canadian province of British Columbia (BC) is in a unique position to contribute to this literature. BC implemented incentive payments targeting chronic disease management within the province-wide fee-for-service system (serving a population of approximately 4.5 million), and with no concurrent changes to that payment system (such as salaried or capitation-based remuneration), to the delivery model (such as teams or medical homes), or to quality measurement or reporting requirements [14]. Imple-

https://doi.org/10.1016/j.healthpol.2017.11.001 0168-8510/© 2017 Published by Elsevier Ireland Ltd.

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model HEAP-3816; No. of Pages 8

ARTICLE IN PRESS M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

2

mentation was staggered over time, with incentives introduced for management of diabetes in 2003, hypertension in 2006, and COPD in 2009. A commissioned evaluation examined cross-sectional comparisons of patients who did and did not have incentives billed for their care and reported that, on balance, the incentive payments corresponded with improved access and continuity, and reduced hospitalizations and health care spending [15,16]. Findings are likely subject to selection bias, as in other research where participation in the intervention is optional and only cross-sectional results are reported [2,4–7]. We used linked administrative data and a quasi-experimental interrupted time series design [17] to produce less biased estimates of the impact of these incentive payments. Due to the fact that incentives were implemented province-wide, we could not construct a sound control group, but staggered introduction of incentives by condition allows us to determine if similar effects are observed for incentives introduced at different points in time. This incentive program is based, in part, on the idea that incentive payments would improve primary care access and continuity for patients with chronic disease, and encourage walk-in clinics to offer more longitudinal and less episodic care [16]. Improved primary care access and continuity would, in was hoped, contribute to enhanced chronic disease management and possibly reduced spending [16,18]. While some patients require hospital care even with high-quality primary care, a goal of chronic disease management in primary care is to prevent acute events, or manage crises in the community, where possible. It is hoped that better primary and secondary preventive care will reduce need for high cost services, and therefore overall health care spending [18,19]. Cross-sectional comparisons found patients with incentives in BC had fewer hospitalizations and lower spending than those without (with the exception of diabetes patients, where hospitalizations were lower but spending higher) [15]. We tracked primary care visits and continuity, process measures of testing and pharmaceutical dispensing, hospitalization rates, and cost of care, before and after the introduction of incentive payments for all patients with diabetes, hypertension, and COPD in BC. These outcome measures reflect care processes mentioned in flow sheets (diabetes and hypertension) or care plan templates (COPD) accompanying incentives, and/or outcomes reported in cross-sectional evaluation of this intervention [15].

2. Methods 2.1. Study context and intervention All physician and hospital services are publically funded under BC’s single-payer Medical Services Plan (MSP), with no out-ofpocket payments or private insurance for medically necessary services. With few exceptions, primary care physicians are paid feefor-service. Province-wide fee codes are negotiated between the BC Ministry of Health and Doctors of BC (called the BC Medical Association before 2014). There is no formal rostering of patients, and no province-wide policy mechanisms to support group or teambased care, nor to pay nurse practitioners, physician assistants, or other non-physician primary care providers. Primary care physicians are expected to coordinate patient care and act as gatekeepers to specialist care. All primary care physicians were eligible to bill for annual payments, in addition to regular visit fees, for providing guideline informed care for patients with diabetes, hypertension, and COPD, over the course of one year. Incentives were introduced in September 2003, April 2006, and September 2009, respectively. Payments were $75 for diabetes (later increased to $125), $50 for hypertension, and $125 for COPD, and are payable once per year,

per patient, through the billing of new fee codes introduced for each condition. Claims for these fee codes include unique physician and patient identifiers which allowed patients for whom incentives were and were not billed to be identified retrospectively. The charts of patients must include documentation of relevant guideline indicated processes of care [16] and flow sheets or care plan templates for each condition were made available as part of billing guides for this purpose [20]. However, charts were not routinely audited, and there was no new measurement or reporting of quality indicators. Following introduction of the COPD incentive in 2009-10, annual spending on these three fee items exceeded $35 million, or over 3% of all fee-for-service payments to primary care physicians in BC. Other incentive programs bring total physician income from incentive payments to over 10%. Clinicians were actively involved in program design as this program was implemented through a partnership between Doctors of BC and the Ministry of Health. Provider surveys conducted in 2010, just after the implementation of the last of the three incentives analyzed (COPD), reported 95% of primary care physicians supported this policy approach [21]. 2.2. Data and study population We used linked, de-identified data developed by the BC Ministry of Health and provided through Population Data BC [22] covering the period from April 2001 to March 2012. The Medical Services Plan (MSP) registration file includes a record for all BC residents who receive or are eligible to receive publicly-funded health care services, including descriptive information about individuals’ age, sex, Health Authority of residence (5 in BC), and number of days in each year registered for health insurance [23]. The MSP payment file includes data on all fee-for-service medical service claims paid to physicians. It describes services billed, including the incentive payments, and includes a patient diagnosis code for each service [24]. The Hospital Separations file includes records of all inpatient and surgical day care discharges and deaths for BC residents, including hospitalizations in other provinces [25]. Each record contains a Resource Intensity Weight variable that can be used to estimate spending. PharmaNet records all prescriptions dispensed in BC, including amount paid, Drug Identification Numbers, and Anatomical Therapeutic Chemical codes [26]. All inferences, opinions, and conclusions drawn in this article are those of the authors, and do not reflect the opinions or policies of the Data Stewards. For each incentive program we examined a period of two years before and two years after the date of introduction. We identified patients qualifying for each program based on ICD codes associated with two outpatient physician visits or one hospitalization during the study period (Supplementary Table 1), following validated algorithms for identifying chronic disease [27]. We excluded individuals who moved into or out of the province over the study period, and who received care from primary care providers not paid fee-forservice, as these patients’ service use is not completely captured in our data. We examined a closed cohort comprised of qualifying patients registered throughout the study period or up until death, regardless of whether or not an incentive was billed for their care. This provides an estimate of the total population-wide effect of incentive payments. 2.3. Outcomes 2.3.1. Primary care use and continuity We tracked the number of visits with any primary care physician (unique combinations of patient/physician/date, regardless of the number of services billed) as a measure of use. Each month, patients were assigned a Usual Provider of Care (UPC) defined as the physician providing the highest number of visits over the preceding year, on a rolling basis. Continuity was measured as the percent of pri-

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model HEAP-3816; No. of Pages 8

ARTICLE IN PRESS M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

mary care visits across the whole study population in each month that occurred with patients’ usual providers of care assigned over the preceding year. 2.3.2. Testing and pharmaceutical dispensing Prescriptions dispensed were identified using anatomic therapeutic chemical (ATC) codes included in province-wide Pharmanet data [26], and all are captured regardless of payer (public, private or out-of-pocket). Testing was measured using laboratory billing codes in the MSP data [24]. We can measure only dispensed prescriptions and completed tests, not prescriptions or laboratory requisitions written. 2.3.3. Hospitalizations Three related variables reflecting hospital use were examined: (1) all acute admissions, (2) hospital admissions through the emergency department, and (3) admissions for selected conditions: acute myocardial infarction, stroke and heart failure among hypertension patients, and COPD. 2.3.4. Health care spending Physician spending is based on billing data, and hospital spending is derived using resource intensity weights for each separation, which assign an average cost based on patient age, sex, diagnoses, and procedures and interventions delivered in hospital, based on values provided by the Canadian Institute of Health Information [28]. Spending on primary care physicians includes the cost of the incentives. Pharmaceutical spending reflects total public, private (insurance), and out-of-pocket payments. We tracked total constant dollar spending, and also examined patterns across the following categories: primary care physicians, medical specialists, surgical specialists, laboratory services, imaging, pharmaceutical use, acute care, and day surgery (Supplementary Fig. 2). 2.4. Design and analysis We used a quasi-experimental interrupted time series design to account for unobservable (but time-invariant) patient characteristics, as well as secular trends surrounding the implementation of the incentives. Staggered introduction of incentives by condition allows us to determine if similar effects are observed for incentives introduced at different points in time. Segmented linear regression models with coefficients for the baseline intercept, pre-intervention trend, level change at time of intervention, and post-intervention change in trend, estimate any immediate changes in the level or longer term trend of chosen outcomes following the introduction of each incentive, while controlling for pre-existing level and trend [29]. Because analysis includes the same individuals followed over time, and not all included individuals were diagnosed with qualifying conditions prior to the study period, we expected health status to decline gradually, and measures of primary care use, testing, prescribing, hospitalizations, and spending to increase over time, absent any effect of incentives [30–32]. This is captured by the terms for pre-intervention trend in each model. Estimates assume that in the absence of the incentive payments pre-intervention trends would have continued. We know of no external factors that would have had a major influence on outcomes, and that changed around the same time as the intervention in the three time periods examined. Data were deseasonalized prior to analysis by decomposing the data into trend, seasonal, and noise components, and subtracting the seasonal component. We checked the autocorrelation and partial-autocorrelation functions (ACF and PACF) for the residuals from ordinary least squares (OLS) regression, and determined the autocorrelation form of the autoregressive–moving-average

3

ARMA(p, q) model for the stationary series. Then, we specified the generalized least squares (GLS) model for error autocorrelation by including the significant autoregressive parameters (p, q). The overall average annual change in each outcome was estimated as the difference in area between the counterfactual (the pre-intervention intercept and trend projected forward) and the observed postintervention values over the two year follow-up period, divided by two, with bootstrapped 95% confidence intervals [33]. To facilitate comparison of the magnitude of effects across outcomes, we also present relative (%) change as the difference in area under the projected and observed lines, divided by the area under the projected line, multiplied by 100. We used SAS Version 9.3 for data preparation and R 3.0.3 for regression analysis (nlme and car packages). Complete model estimates are reported in Supplementary Table 2. We conducted sensitivity analyses to confirm consistent results are obtained using different study designs. We examined a cohort of only patients who received incentives, with time zero set as the date of individual incentive billing, not the calendar date of incentive introduction. We also assessed whether patterns observed in the 4year study periods and constant study cohorts reported here were consistent with annual patterns from 2001/2 to 2011/12 among all patients with recorded diagnoses for each condition, in each year. We considered analysis with control, however, given all patients were eligible, selection processes that determined which patients received incentives for their care made it difficult to construct methodologically sound control groups. The incentive programs themselves appear to have changed diagnostic patterns for these conditions [34]. Given more frequent diagnoses after incentive introduction, patients with first recorded diagnoses before and after the programs likely differ in ways we cannot measure. Even among patients with pre-existing disease, contact with the healthcare system in the period after incentive introduction is both a precondition to receiving the intervention, and an outcome of interest. All else equal, patients with more frequent contact with primary care in the post-intervention period were more likely to receive an incentive. Given these considerations, and the fact that this was a population-based intervention, analysis of the entire eligible population was both the most rigorous and appropriate choice.

3. Results We identified 189,583 patients with diabetes, 507,030 patients with hypertension, and 82,497 patients with COPD who were registered for the provincial health insurance scheme over the relevant study periods and received fee-for-service primary care (Table 1). Of these, between 41.9% (hypertension) and 45.3% (COPD) had at least one incentive billed for their care following incentive introduction. Within our study 46.1% of diabetes patients, 53.4% of hypertension patients, and 48.2% of COPD patients were female (Table 1). Patients with incentives billed were slightly less likely to be female among diabetes patients, but more likely to be female among hypertension and COPD patients. Patients with incentives billed were more likely to be ages 45–74 than older (75+) or younger (<45 years). In general patients with these three chronic conditions are disproportionately represented in lower income quintiles, but patients in the highest three income quintiles were slightly more likely to have incentives billed. Patients with COPD had the most chronic conditions, followed by diabetes and hypertension. Hypertension and diabetes patients with incentives had fewer chronic conditions than patients without incentives, whereas COPD patients with incentives had slightly more chronic conditions than those without. For all conditions, patients who received incentives were less likely to have had a recorded qualifying diagnosis in the

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model HEAP-3816; No. of Pages 8

ARTICLE IN PRESS M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

4

Table 1 Characteristics of British Columbia chronic disease patients within the study period, N(%). Diabetes

Hypertension

COPD

All patients

Patients with incentives

All patients

Patients with incentives

All patients

Patients with incentives

N

189,583 (100)

79,939 (42.2)

507,030 (100)

212,502 (41.9)

82,497 (100)

37,361 (45.3)

Sex Female Male

87,337 (46.1) 102,098 (53.9)

36,345 (45.5) 43,536 (54.5)

270,602 (53.4) 236,063 (46.6)

117,961 (55.5) 94,406 (44.4)

39,790 (48.2) 42,670 (51.7)

18,108 (48.5) 19,235 (51.5)

Age <45 years 45–74 years 75+ years

25,117 (13.2) 130,194 (68.7) 34,272 (18.1)

8893 (11.1) 58,208 (72.8) 12,838 (16.1)

49,649 (9.8) 347,097 (68.5) 110,284 (21.8)

18,118 (8.5) 151,275 (71.2) 43,109 (20.3)

5313 (6.4) 50,928 (61.7) 26,256 (31.8)

1898 (5.1) 25,135 (67.3) 10,328 (27.6)

Health Authority Interior Fraser Vancouver Coastal Island Northern

34,534 (18.3) 66,915 (35.4) 46,184 (24.4) 29,587 (15.6) 11,892 (6.3)

16,352 (20.5) 27,540 (34.5) 17,621 (22.1) 12,676 (15.9) 5540 (6.9)

114,603 (22.6) 182,830 (36.1) 102,435 (20.2) 76,337 (15.1) 30,250 (6.0)

49,070 (23.1) 73,351 (34.6) 44,023 (20.7) 34,797(16.4) 11,029 (5.2)

15,721 (26.1) 18,913 (31.4) 12,688 (21.1) 9376 (15.6) 3515 (5.8)

2196 (25.4) 2921 (33.7) 1810 (20.9) 1402 (16.2) 332 (3.8)

Income quintile 1 (Lowest) 2 3 4 5 (Highest)

42,692 (24.1) 38,414 (21.7) 35,237 (19.9) 31,930 (18.0) 29,128 (16.4)

17,934 (23.2) 38,414 (21.4) 35,237 (20.1) 31,930 (18.7) 29,128 (16.7)

100,656 (20.4) 103,577 (21.0) 100,383 (20.4) 96,596 (19.6) 91,306 (18.5)

40,543 (19.3) 42,533 (20.2) 43,270 (20.5) 42,359 (20.1) 41,932 (19.9)

19,912 (25.8) 17,148 (22.2) 15,519 (20.1) 13,689 (17.7) 10,955 (14.2)

9366 (25.4) 8159 (22.1) 7482 (20.3) 6548 (17.7) 5386 (14.6)

Number of chronic conditions (at the end of the study period) 57,190 (30.2) 23,236 (29.1) 1 69,025 (36.4) 31,727 (39.7) 2 50,416 (26.6) 20,931 (26.2) 3–4 12,952 (6.8) 4045 (5.1) 5+

238,088 (47) 135,717 (26.8) 107,644 (21.2) 25,581 (5.0)

121,696 (57.3) 51,534 (24.3) 34,035 (16) 5237 (2.5)

9540 (11.6) 14,385 (17.4) 28,740 (34.9) 29,832 (36.2)

3885 (10.4) 6417 (17.2) 13,073 (35) 13,986 (37.4)

Number of incentives billed in the study period 0 109,644 (57.8) 0 (0) 45,675 (24.1) 45,675 (57.1) 1 34,264 (18.1) 34,264 (42.9) 2+

294,528 (58.1) 116,973 (23.1) 95,529 (18.8)

0 (0) 116,973 (55.0) 95,529 (45.0)

45,136 (54.7) 21,111 (25.6) 16,250 (19.7)

0 (0) 21,111 (56.5) 16,250 (43.5)

Qualifying disease Prior to study First 24 months Second 24 months Died

305,307 (60.2) 98,666 (19.5) 103,057 (20.3) 25,886 (5.1)

137,278 (64.6) 33,095 (15.6) 42,129 (19.8) 2754 (1.3)

26,004 (31.5) 23,413 (28.4) 33,080 (40.1) 13,286 (16.1)

11,153 (29.9) 7165 (19.2) 19,043 (51) 2179 (5.8)

100,363 (52.9) 44,019 (23.2) 45,201 (23.8) 17,626 (9.3)

46,151 (57.7) 14,800 (18.5) 18,988 (23.8) 2299 (2.9)

Note: Missing sex for 148 diabetes patients, 365 hypertension patients, and 37 COPD patients. Missing HA for 471 diabetes patients, 575 hypertension patients, and 114 COPD patients. 13 diabetes patients, 90 hypertension patients, and 3 COPD patients had 3 or more incentives billed in the study period. Missing income quintile for 12,182 diabetes patients, 14,512 hypertension patients and 5274 COPD patients. Patients with incentives significantly different from those without incentives for all covariates (p < 0.05).

24 months preceding incentive introduction. Recorded diagnoses increased following incentive introduction, and billing of incentives often coincided with first-time recorded diagnoses. This reflects the fact that diagnoses observed in administrative data may not reflect the true first date that conditions were diagnosed, and that incentives may have motivated changes in coding practice. A lower proportion of deaths among patients with incentives was both expected and observed (Table 1). This is an artifact of study design, in that patients must have survived to time 0 in order to receive the incentive. Death rates were constant over time within the study cohorts (data not shown) and so any effect on outcome variables is captured in the pre-intervention trend. At study outset there were between 74.4 (hypertension) and 105 (COPD) visits with primary care physicians per 100 patients each month and between 73.2% (COPD) and 78.9% (diabetes) of these were with the usual provider of care (Supplementary Table 2). Following incentive introduction, there was a statistically significant but likely clinically unimportant decline in primary care visits with the patients’ usual provider of care from 77.6% to 75.3% among patients with hypertention, and there were no other notable changes in primary care visits or continuity (Table 2). Increases in primary care visits over time reflect the progression of disease in a closed cohort and do not reflect secular changes among all patients in BC.

Absent any effect of the incentives we would expect testing to increase over the study period at a constant slope, but in several cases testing and prescribing increased more rapidly that would have been predicted by pre-incentive trends (Fig. 1). Diabetes flow sheets suggested HbA1c and fasting glucose every 3–6 months. We observe a small increase in trend following incentive introduction, leading to a small but not statsistically significant increase relative to pre-incentive trends projected forward (approximately 5%) (Table 2). At the conclusion of the study period patients in the diabetes cohort were receiving plasma glucose and HbA1c less than twice annually, on average. Both diabetes and hypertenion flow sheets suggest ACR, eGFR and lipids annually or as otherwise indicated. Prior to incentive introduction members of the these cohorts received eGFR tests roughtly annually, though ACR tests and full lipids panels were less frequent (we count all three lipids tests individually, not as a panel). For both ACR and lipids we observe level increases at the time of incentive introduction, though average changes only reached statistical significance for ACR testing among patients with hypertension (Table 2), for whom testing increased by 37.9% relative to pre-incentive trends. At the conclusion of the study period patients were still receiving ACR testing less than annually. Spirometry is used to diagnose COPD but not as part of regualar monitoring. Rates are correspondingly much lower. A

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model

ARTICLE IN PRESS

HEAP-3816; No. of Pages 8

M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

5

Table 2 Estimated average annual change (95%CI) and percent change for study outcomes. Diabetes

%

Hypertension

%

COPD

%

Primary care Visits with primary care physicians (per 100 patients) Continuity of carea

−42.9 (−99.8, 14.0) 0.32 (−0.54, 0.14)

−3.8 0.42

−7.3 (−66.9, 52.2) −2.3 (−3.2, −0.15)

−0.8 −2.9

40.7 (−82.2, 90.4) −0.42 (−1.6, 0.092)

−0.3 −0.56

Lab/diagnostic tests (per 100 patients) HbA1c Fasting glucose Albumin/creatinine ratio (ACR) Estimated Glomerular Filtration Rate (eGFR) Lipids (TC/HDL, LDL-C, Triglycerides) Spirometry

8.7 (−0.60, 7.9) 6.1 (−2.4, 14.6) 4.3 (−0.86, 9.4) 1.5 (−7.5, 10.5) 17.7 (−1.7, 37.2) –

5.7 4.9 6.7 1.1 6.7

– 3.8 (−3.7, 11.3) 10.4 (5.5, 15.3) 5.1 (−3.9, 14.1) 17.6 (−3.0, 38.2) –

4.4 37.9 4.4 8.6

– – – – – 3.9 (−1.8, 9.6)

9.9

Prescriptions (per 100 patients) Insulin and antihyperglycemics Antihypertensives Beta2 agonist Inhaled corticosteroid Prednisone Antibiotic

3.85 (−2.20, 2.98) – – – – –

0.7

– – 10.9 (−1.5, 23.3) 2.6 (−1.2, 6.4) 1.1 (−2.3, 4.5) 0.3 (−1.2, 1.8)

3.8 3.3 2.1 1.4

−4.9 (−19.7, 9.8) −6.9 (−16.0, 2.2)

−2.3 −5.0

8.9 (−12.6, 30.5) 7.2 (−8.7, 23.1)

2.5 2.8

14.0 (2.3, 25.8)

18.8

130 (−320, 580)

1.6

Hospitalizations (per 1000 patients) Total hospitalizations Hospitalizations via emergency department AMI Stroke Heart failure COPD Health care costs (per patient) Total expenditures (CAD)

−180 (−450, 88)

−3.2

– 62.5 (35.2, 89.8) – – – – −5.0 (−12.5, 2.4) −4.1 (−8.8, 0.67) −0.19 (−1.13, 0.75) −0.56 (−0.98, −0.15) −1.41 (−2.54, −0.27)

260 (73, 440)

8.3

−3.3 −4.6 −2.7 −12.3 −12.0

7.0

Note: The average annual change is the difference in area between projected pre-intervention trend line and observed post-intervention fitted line over the 24 months of follow-up, divided by two. Percent change is the difference in area under the projected and observed lines, divided by the area under the projected line, multiplied by 100. Estimates significant at p ≤ 0.05 indicated in bold. Values rounded to display a minimum of two significant figures. a Continuity of care estimates reflect the average change in the percent of primary care visits that occurred with patients’ usual provider of care (UPC) between the post-incentive period and pre-intervention period.

Fig. 1. Laboratory/diagnostic tests and prescriptions dispensed per 100 patients.

pronounced level change was observed immediately following incentive introduction (Fig. 1), though the trend declined thereafter, and the average annual change did not reach statistical significance (Table 2). Changes in prescribing were also variable. We observed no increase in prescriptions dispensed for antihyperglycemics and insulin, while antihypertensives increased significantly, with 62.5 additional prescriptions per 100 patients each year, a relative increase of over 8%. While significant level increases for beta2 agonists and inhaled corticosteroids dispensed to patients with COPD were observed (Supplementary Table 2), average annual changes in these measures were not significant (Table 2).

We observe no significant changes in total hospitalizations or hospitalizations via the emergency department (Table 2, Fig. 2). Patterns of hospitalization for selected conditions are variable: we observed 12.3% and 12.0% decreases in hospitalizations for stroke and heart failure among hypertention patients and a 18.8% increase in hospitalizations for COPD (Table 2). Rates of hospitalization for these selected conditions are less stable over time than total hospializations, so these results should be interpreted with caution. Patterns of health care spending also vary by condition (Supplementary Fig. 2). A significant increase in spending among hypertension patients (Table 2) reflects additional payments to GPs, but also increased spending on pharmaceuticals (Supplementary

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model HEAP-3816; No. of Pages 8

ARTICLE IN PRESS M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

6

Fig. 2. Hospitalization rates per 1000 patients.

Fig. 2b). Spending on primary care physician services also increased among diabetes patients, but this was counterbalanced by declines in acute care spending such that the overal change was not significant. There was a marked increase in GP spending among COPD patients, but this represented a small enough proportion of total spending (<10%) given very high spending on acute care within this population, that the overal annual increase in spending did not reach statistical significance (Table 2). Results of sensitivity analyses are consistent with findings reported above, and offer reassurance that meaningful effects among patients who received incentives are not obscured by this population-level analysis (data not shown). As expected, among patients who received incentives for their care, we observed a spike in primary care visits and spending in the month of incentive billing. We also observed a pronounced spike in hospitalizations in the month immediately preceeding incentive billing, as an acute event may have prompted diagnosis or otherwise brought the patient’s eligibility for the incentive to the attention of their primary care provider. Otherwise, trends in primary care use and hospitalizations are consistent before and after incentive billing. There are also spikes in testing and prescribing at time of incentive billing, but rates return to pre-period trends for all but a handful of care processes (HbA1c and lipid panels among diabetes patients, ACR and lipid panels testing hypertension patients, and prednisone dispensing among COPD patients). These observations are consistent with the small increases in level and/or trend observed across the full study population. Glucose tolerance testing and spirometry spike dramatically in the month prior to incentive billing, and then decline rapidly, as would be expected among patients with a confirmed diagnosis at time zero. Finally, examining outcomes across all years for which data are available, rates of laboratory monitoring continued to increase among patients with diabetes and hypertention until 2008/9 but then plateaued or declined in more recent years. Prescribing among COPD patients has been falling over time, though this may in part reflect increasing diagnosis or recording of diagonsis, and more patients with less severe disease reflected in the denominator of annual values. 4. Discussion We examined effects of incentive payments to primary care physicians for the care for patients with diabetes, hypertension, and COPD on visits with primary care physicians, continuity of care, laboratory testing, pharmaceutical dispensing, hospitaliza-

tions, and total health care spending. We observed no increases in primary care visits or continuity. While we observe increases in some measures of testing and prescribing at the time incentives were introduced, total changes in these processes of care over the two years of follow up were largely not statistically significant. We observe no change in total hospitalizations or hospitalizations via the emergency department. Changes in hospitalization rates for selected conditions were inconsistent, with a decrease in hospitalizations for stroke and heart failure among patients with hypertension, but an increase in hospitalizations for COPD. Health care spending increased significantly for patients with hypertension. Our results are generally consistent with previous research that has found some limited impact of incentive payments on primary care process measures [2–9,35–37], but no consistent impact on spending or hospitalizations [3,5,6]. Our findings are not consistent with reduced total hospitalizations and cost savings as was observed in cross-sectional comparisons of patients who did and did not have incentives billed for their care [15]. Our finding of lower rates of hospitalization for stroke and heart failure among patients with hypertension is notable, and may be interpreted as evidence of a successful incentive program in this patient group. We believe reductions in hospitalization rates should be interpreted with caution, given instability in hospitalization rates for selected conditions. More research is needed to understand changes in rates of hospitalizations for heart failure and stroke over time There may be several reasons why we find only limited evidence of impact. First, relatively low uptake (42.2–45.3% of potentiallyeligible patients) is an important consideration. This is in part due to the fact that we include patients with recorded diagnoses at any point in the study period, so some may have only qualified for the intervention after its introduction. However, sensitivity analysis examining only those patients who received incentives shows similarly modest effects. Uptake may also reflect low rates of use of Electronic Medical Records at time of implementation [14], which would permit physicians to identify all patients with qualifying chronic conditions in their panel, rather than only those who presented in office after incentive introduction. Second, these incentive payments together constitute only a fraction of BC primary care physicians’ income (∼3%). For comparison, primary care incentives implemented under the Quality and Outcomes Framework in the UK made up almost a quarter of physician income [38]. That said, small incentives (less than 5% of annual income) have modified practice in some settings [4], and a dose-response relationship has not generally been observed

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model HEAP-3816; No. of Pages 8

ARTICLE IN PRESS M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

[4,7]. Though diabetes and COPD payments were of greater value than hypertension, we observe more significant changes within the hypertension cohort. Third, systematic reviews have reported greater impact on outcomes that have more room for improvement [2,7]. Prior to incentives, patients saw primary care physicians almost monthly and had roughly three-quarters of visits with their usual provider of care (Supplementary Table 2). The only significant change among laboratory testing measured was for ACR testing among patients with hypertension, which had among the lowest pre-incentive rates, and which was recommended annually in the flowsheet provided to clinicians. At time of incentive introduction HbA1c, glucose, eGFR, and lipid were closer to the recommended frequency, so it is plausible there was more limited room for change. Fourth, this intervention lacked the systematic tracking and reporting of quality indicators central to pay-for-performance initiatives such as the UK’s Quality and Outcomes Framework, which has shown some more positive effects [12]. It focused on payments to individual physicians, and lacked components of the Chronic Care Model seen in other successful interventions and high-performing primary care systems, such as enhanced clinical information systems and team-based care [39–42]. Our study has several limitations that warrant discussion. This study was designed retrospectively and we are limited to outcome measures available using administrative data. We have no information on the results of laboratory testing, nor any other objective measures of disease severity. We cannot determine if testing and prescribing is appropriate based on individual patient characteristics, and we are unable to measure prescriptions and laboratory requisitions written, only tests performed and medications dispensed. Our intention in reporting these frequencies is not to inform quality of care directly, but simply to identify any changes in care process that might plausibly result from incentive introduction. It is also possible that in aggregate processes of care remained unchanged, but these were better targeted at the individual level (i.e. increases in appropriate testing and prescribing may have been negated by decreases in inappropriate testing and prescribing). We were also unable to measure counseling for lifestyle management and patient education, nor could we measure eye exams among diabetic patients, influenza and pneumococcus immunizations, and oxygen therapy among COPD patients. In addition to limited measures, were are unable to determine the extent to which it is the payments, or simply the provision of flow sheets and care plan templates at the time of incentives that is responsible for increases in testing and prescribing. Finally, we are only able to examine relatively short-term impacts of the incentive program. Hospitalization rates are at best an incomplete proxy for the impact of care on disease progression, though likely reflect problems of access to community-based care more immediately. In addition, any impact on hospitalization rates requires that incentives have an impact on physician behaviour, and that physician behaviour has an impact on hospitalization rates. Whether the latter is plausible likely varies by condition. In the absence of a strong control group, attributing longer-term changes in hospitalizations, mortality, or other measures of health status or disease progression is not possible. An incentive payment for management of congestive heart failure (CHF) was also introduced at the time of the diabetes incentive. However, issues around diagnostic criteria meant that uptake was initially extremely low (∼15% in the first two years) and so we excluded this program from analysis. It may be that there was greater room for improvement in management of CHF in primary care, with more widespread undertreatment at the time of incentive introduction. Results from diabetes, hypertension, and COPD programs may therefore not apply to CHF. In 2007 a separate complex care incentive program targeting patients with two or more

7

conditions was introduced. COPD was included under this program, which may mean that some patients with COPD had already received incentive-based care under that program. However, only the COPD-specific program included a care plan template, and both incentives were payable on the same patient provided criteria for both programs were met. Findings suggest that policies focused on incentive payments alone, absent other interventions, may have only modest impact. Payments tied to measures of health system performance where there is recognized room for improvement, and where there is consistent tracking and reporting, could have greater impact [2,7]. Payment reform has a role to play in supporting high-preforming primary care [42] and strengthening care for patients with chronic illness [41]. However, changes in payments need be tied to other interventions targeting delivery system design (such as linking patients to a medical home or team-based care), and structural supports for clinicians (such as clinical information systems equipped to support population management) [42]. 5. Conclusion We find only limited evidence of impact of incentive payments on the outcome measures examined. Absent other interventions, we observe small changes in patient management and hospitalization rates among patients with hypertension, but no changes in primary care visits and continuity, and no similar reductions in hospitalizations or changes in spending among patients with diabetes and COPD in BC. Policymakers should consider other approaches to modify service delivery models, and support all patients and primary care providers in managing chronic disease. Conflict of interest statement The authors have no conflicts of interest to declare. Acknowledgments This research was funded by Canadian Institutes for Health Research (CIHR) operating grant MOP-126008 “Incentive payments to British Columbia primary care physicians for chronic disease management: What is the effect on patient care?” Dr. Law received salary support through a Canada Research Chair and a Michael Smith Foundation for Health Research Scholar Award. The researchers are independent of the funder, and CIHR had no roll in conducting the study, writing the report, or in the decision to submit this article for publication. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.healthpol.2017.11. 001. References [1] The Commonwealth Fund. International Profiles of Health Care Systems; 2016. p. 2015. http://www.commonwealthfund.org/∼/media/files/publications/ fund-report/2016/jan/1857 mossialos intl profiles 2015 v7.pdf (Accessed 1 November 2016). [2] 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;110:115–30, http://dx.doi.org/10.1016/j.healthpol.2013.01.008. [3] Emmert M, Eijkenaar F, Kemter H, Esslinger AS, Schöffski O. Economic evaluation of pay-for-performance in health care: a systematic review. European Journal of Health Economics 2012;13:755–67, http://dx.doi.org/10.1007/s10198-011-0329-8. [4] Houle SK, McAlister FA, Jackevicius CA, Chuck AW, Tsuyuki RT. Does performance-based remuneration for individual

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001

G Model HEAP-3816; No. of Pages 8

ARTICLE IN PRESS M.R. Lavergne et al. / Health Policy xxx (2017) xxx–xxx

8

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21] [22]

health care practitioners affect patient care? A systemreview. Annals of Internal Medicine 2012;157:889–99, atic http://dx.doi.org/10.7326/0003-4819-157-12-201212180-00009. Scott A, Sivey P, Ait Ouakrim D, Willenberg L, Naccarella L, Furler J, et al. The effect of financial incentives on the quality of health care provided by primary care physicians (review). Cochrane Library 2011:9. Flodgren G, Eccles MP, Shepperd S, Scott A, Parmelli E, Beyer FR. An overview of reviews evaluating the effectiveness of financial incentives in changing healthcare professional behaviours and patient outcomes. Cochrane Database of Systematic Reviews 2011:CD009255, http://dx.doi.org/10.1002/14651858.CD009255. Van Herck P, De Smedt D, Annemans L, Remmen R, Rosenthal MB, Sermeus W. Systematic review: effects, design choices, and context of pay-forperformance in health care. BMC Health Services Research 2010;10:247, http://dx.doi.org/10.1186/1472-6963-10-247. Christianson JB, Leatherman S, Sutherland K. Lessons from evaluations of purchaser pay-for-performance programs: a review of the evidence. Medical Care Research and Review 2008;65:5S–35S, http://dx.doi.org/10.1177/1077558708324236. Rosenthal MB, Frank RG. What is the empirical basis for paying for quality in health care? Medical Care Research and Review 2006;63:135–57, http://dx.doi.org/10.1177/1077558705285291. Fagan PJ, Schuster AB, Boyd C, Marsteller JA, Griswold M, Murphy SME, et al. Chronic care improvement in primary care: evaluation of an integrated pay-for-performance and practice-based care coordination program among elderly patients with diabetes. Health Services Research 2010;45:1763–82, http://dx.doi.org/10.1111/j.1475-6773.2010.01166.x. Petersen LA, Ramos KS, Pietz K, Woodard LD. Impact of a pay-for-performance program on care for black patients with hypertension: important answers in the era of the affordable care act. Health Services Research 2016:1–18, http://dx.doi.org/10.1111/1475-6773.12517. Harrison MJ, Dusheiko M, Sutton M, Gravelle H, Doran T, Roland M. Effect of a national primary care pay for performance scheme on emergency hospital admissions for ambulatory care sensitive conditions: controlled longitudinal study. BMJ 2014;349:g6423, http://dx.doi.org/10.1136/bmj.g6423. Kontopantelis E, Springate D, Reeves D, Ashcroft DM, Valderas JM, Doran T. Withdrawing performance indicators: retrospective analysis of general practice performance under UK Quality and Outcomes Framework. BMJ 2014;348:g330, http://dx.doi.org/10.1136/bmj.g330. Hutchison B, Levesque J-F, Strumpf E, Coyle N. Primary health care in Canada: systems in motion. Milbank Quarterly 2011;89:529–32, http://dx.doi.org/10.1111/j.1468-0009.2008.00538.x. Hollander M, Kadlec H. Incentive-based primary care: cost utilization analysis. Permanente Journal 2015;19:1–11, and http://dx.doi.org/10.7812/TPP/15-045. Mazowita G, Cavers WJ. Issues in international health policy: reviving full-service family practice in British Columbia. Commonwealth Fund 2011;1538:1–20. Kontopantelis E, Doran T, Springate DA, Buchan I, Reeves D. Regression based quasi-experimental approach when randomisation is not an option: interrupted time series analysis. BMJ 2015;350:h2750, http://dx.doi.org/10.1136/bmj.h2750. Hollander MMJ, Kadlec H, Hamdi R. Increasing value for money in the canadian healthcare system: new findings on the contribution of primary care services. Healthcare Quarterly 2009;12:32–45. Haggerty JL, Reid RJ, Freeman GK, Starfield BH, Adair CE, McKendry R. Continuity of care: a multidisciplinary review. British Medical Journal 2003;327:1219–21, http://dx.doi.org/10.1136/bmj.327.7425.1219. General Practice Services Committee. Chronic Disease Management Incen2012. http://www.gpscbc.ca/system/files/GPSC Chronic Disease tives Management Billing Guide revised January 2012.pdf (Accessed 27 July 2015). Tregillus VHF, Cavers WJ. General practice services committee: improving primary care for BC physicians and patients. Healthcare Quarterly 2011;14:1–6. Pencarrick Hertzman C, Meagher N, McGrail KM. Privacy by Design at Population Data BC: a case study describing the technical, administrative, and

[23] [24] [25] [26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

[34]

[35]

[36]

[37]

[38]

[39]

[40]

[41] [42]

physical controls for privacy-sensitive secondary use of personal information for research in the public interest. Journal of the American Medical Informatics Association 2012:25–8, http://dx.doi.org/10.1136/amiajnl-2012-001011. British Columbia Ministry of Health. Consolidation file (MSP registration & premium billing). Population Data BC; 2011. Data extract. MOH. British Columbia Ministry of Health. Medical services plan (MSP) payment information file. Population Data BC; 2011. Data extract. MOH. CIHI. Discharge abstract database (hospital separations). Population Data BC; 2011. Data Extract. MOH. BC Ministry of Health, 2011, PharmaNet. V2. BC Ministry of Health [publisher]. Data Extract. Data Stewardship Committee (2011). http://www.popdata.bc.ca/ data. Lix L, Yogendran M, Burchill C, Metge C, Mckeen N, Moore D, et al. Defining and validating chronic diseases: an administrative data approach. Winnipeg: Manitoba Centre for Health Policy; 2006. CIHI. Canadian Hospital Reporting Project Technical Notes – Clinical Indicators; 2013. http://publications.gc.ca/collections/collection 2013/icis-cihi/H118-861-2013-eng.pdf (Accessed 10 November 2015). Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time series studies in medication use research. Journal of Clinical Pharmacy and Therapeutics 2002;27:299–309. The action to control cardiovascular risk in diabetes study group. Effects of intensive glucose lowerin in type 2 diabetes. New England Journal 2008;358:2545–59, http://dx.doi.org/10.1056/NEJMoa1610528. Upadhya B, Rocco M, Lewis CE, Oparil S, Lovato LC, Cushman WC, et al. Effect of intensive blood pressure treatment on heart failure events in the systolic blood pressure reduction intervention trial. Circulation Heart Failure 2017;10:1–11, http://dx.doi.org/10.1161/CIRCHEARTFAILURE.116.003613. Vogelmeier D, Herderer B, Glaab T, Schmidt H, Rutten-van Molken M, Beeh K, et al. Tiotropium versus salmeterol for the prevention of exacerbations of COPD. New England Journal 2011;364:1093–103, http://dx.doi.org/10.1056/NEJMoa0904327. Zhang F, Wagner AK, Soumerai SB, Ross-Degnan D. Methods for estimating confidence intervals in interrupted time series analyses of health interventions. Journal of Clinical Epidemiology 2009;6:2, http://dx.doi.org/10.1016/j.biotechadv.2011.08.021.Secreted. McGrail K, Lavergne R, Lewis S. The chronic disease explosion: artificial bang or empirical whimper? BMJ 2016;1312:i1312, http://dx.doi.org/10.1136/bmj.i1312. Li J, Hurley J, Decicca P, Buckley G. Physician response to pay-for-performance: evidence from a natural experiment. Health Economics 2014;23:962–78, http://dx.doi.org/10.1002/hec. Kiran T, Wilton AS, Moineddin R, Paszat L, Glazier RH. Effect of payment incentives on cancer screening in Ontario primary care. Annals of Family Medicine 2014;12:317–23, http://dx.doi.org/10.1370/afm.1664. Kiran T, Victor JC, Kopp A, Shah BR, Glazier RH. The relationship between financial incentives and quality of diabetes care in Ontario, Canada. Diabetes Care 2012;35:1038–46, http://dx.doi.org/10.2337/dc11-1402. Campbell S, Reeves D, Kontopantelis E, Sibbald B, Roland M. Effects of pay for performance on the quality of primary care in england. New England Journal 2009;361:368–78, http://dx.doi.org/10.1056/NEJMsa0807651. Zwerink M, Brusse-Keizer M, van Der Valk PD, Zielhuis GA, Monninkhof EM, van Der Palen J, et al. Self management for patients with chronic obstructive pulmonary disease. Cochrane Database of Systematic Reviews 2014;19:CD002990, http://dx.doi.org/10.1002/14651858.CD002990.pub3. Adams SG, Smith PK, Allan PF, Anzueto A, Pugh JA, Cornell JE. Systematic review of the chronic care model in chronic obstructive pulmonary disease prevention and management. Archives of Internal Medicine 2007;167:551–61, http://dx.doi.org/10.1001/archinte.167.6.551. Bodenheimer T, Wagner EH, Grumbach K. Improving primary care for patients with chronic illness. JAMA 2002;288:1775–9. Bodenheimer T, Ghorob A, Willard-Grace R, Grumbach K. The 10 building blocks of high-performing primary care. Annals of Family Medicine 2014;12:166–71, http://dx.doi.org/10.1370/afm.1616.

Please cite this article in press as: Lavergne MR, et al. Effect of incentive payments on chronic disease management and health services use in British Columbia, Canada: Interrupted time series analysis. Health Policy (2017), https://doi.org/10.1016/j.healthpol.2017.11.001