Estimating the Burden of Osteoarthritis to Plan ... - Wiley Online Library

Download PDF
23 downloads 13513 Views 200KB Size Report
Comment
Resource planning to meet this increasing need requires esti- mates of the anticipated .... expertise in administrative data analysis (CB and DM). Incidence and ...
Arthritis Care & Research Vol. 67, No. 10, October 2015, pp 1379–1386 DOI 10.1002/acr.22612 C 2015, American College of Rheumatology V

ORIGINAL ARTICLE

Estimating the Burden of Osteoarthritis to Plan for the Future DEBORAH A. MARSHALL,1 SONIA VANDERBY,2 CHERYL BARNABE,1 KAREN V. MacDONALD,1 COLLEEN MAXWELL,3 DIANNE MOSHER,1 TRACY WASYLAK,4 LISA LIX,5 ED ENNS,6 CY FRANK,7 AND TOM NOSEWORTHY8

Objective. With aging and obesity trends, the incidence and prevalence of osteoarthritis (OA) is expected to rise in Canada, increasing the demand for health resources. Resource planning to meet this increasing need requires estimates of the anticipated number of OA patients. Using administrative data from Alberta, we estimated OA incidence and prevalence rates and examined their sensitivity to alternative case definitions. Methods. We identified cases in a linked data set spanning 1993 to 2010 (population registry, Discharge Abstract Database, physician claims, Ambulatory Care Classification System, and prescription drug data) using diagnostic codes and drug identification numbers. In the base case, incident cases were captured for patients with an OA diagnostic code for at least 2 physician visits within 2 years or any hospital admission. Seven alternative case definitions were applied and compared. Results. Age- and sex-standardized incidence and prevalence rates were estimated to be 8.6 and 80.3 cases per 1,000 population, respectively, in the base case. Physician claims data alone captured 88% of OA cases. Prevalence rate estimates required 15 years of longitudinal data to plateau. Compared to the base case, estimates are sensitive to alternative case definitions. Conclusion. Administrative databases are a key source for estimating the burden and epidemiologic trends of chronic diseases such as OA in Canada. Despite their limitations, these data provide valuable information for estimating disease burden and planning health services. Estimates of OA are mostly defined through physician claims data and require a long period of longitudinal data.

Osteoarthritis (OA), the most common form of arthritis, is a disabling chronic disease with significant clinical and economic implications (1–5). Key OA risk factors include age, sex, obesity, and trauma (6–9). As the Canadian population is aging and obesity is increasing, OA incidence and prevalence are expected to continue rising (4,5,7–10). The Arthritis Alliance of Canada reported that 25% of Canadians and almost 30% of the labor force are expected

to have OA by 2040 (11). Yet substantial uncertainty surrounds projections of OA incidence and prevalence rates, partly due to varying OA definitions and how incidence and prevalence are calculated (2–5). Incidence and prevalence are valuable for measuring disease burden and future health care needs. As OA is a chronic disease, an incident case remains a prevalent case for the remainder of the individual’s life. OA incidence and prevalence estimates are often based on self-reported data from national surveys (8,12,13). Self-

Supported by grants from Alberta Innovates Health Solutions, the Canadian Institutes of Health Research, and the Arthritis Society of Canada. Dr. Marshall’s work was supported by a Canada Research Chair in Health Systems and Services Research, and an Arthur J. E. Child Chair in Rheumatology. 1 Deborah A. Marshall, PhD, Cheryl Barnabe, MD, MSc, FRCPC, Karen V. MacDonald, MPH, Dianne Mosher, MD, FRCPC: University of Calgary, Calgary, Alberta, Canada; 2 Sonia Vanderby, PhD: University of Saskatchewan, Saskatoon, Saskatchewan, Canada; 3Colleen Maxwell, PhD: University of Waterloo, Waterloo, Ontario, Canada; 4Tracy Wasylak, CHE, MSc, BN: Alberta Health Services, Calgary, Alberta, Canada; 5 Lisa Lix, PhD, PStat: University of Manitoba, Winnipeg,

Manitoba, Canada; 6Ed Enns, MEc: Alberta Bone and Joint Health Institute, Calgary, Alberta, Canada; 7Cy Frank, MD, FRCPC: Alberta Innovates-Health Solutions and University of Calgary, Calgary, Alberta, Canada; 8Tom Noseworthy, MD, MSc, MPH, FRCPC, FACP: Alberta Health Services and University of Calgary, Calgary, Alberta, Canada. Dr. Marshall has received consultancy fees (less than $10,000) from Optum Insight. Address correspondence to Deborah A. Marshall, PhD, University of Calgary, Room 3C56, Health Research Innovation Centre, 3280 Hospital Drive NW, Calgary AB T2N 4Z6, Canada. E-mail: [email protected]. Submitted for publication December 15, 2014; accepted in revised form April 28, 2015.

INTRODUCTION

1379

1380

Significance & Innovations  With aging and obesity trends, the incidence and prevalence of osteoarthritis (OA) are expected to rise, increasing the demand for health resources. Resource planning to meet this increasing need requires estimates of the anticipated number of OA patients.  Using administrative data from Alberta, Canada, this research has demonstrated that incidence and prevalence estimates of OA are mostly defined through physician claims data and are sensitive to alternative case definitions.  Chronic diseases such as OA require a long period of longitudinal data to avoid underestimating prevalence.  Administrative databases are a key source for estimating the burden and epidemiologic trends of chronic diseases such as OA.

reported data are known to overestimate prevalence of a disease (7) and typically only reflect prevalence, not incidence. Further, surveys often ask about “arthritis” and do not report OA-specific results (2). Accurate estimates of disease incidence and prevalence are needed for developing accurate high-quality forecasts of future health resource requirements. Administrative health databases are an efficient data source for population-based estimates and are increasingly being used to obtain estimates of disease burden, quality of care, and resource use (14,15). The work of the Canadian Rheumatology Administrative Data (CANRAD) Network to advance the science of administrative health data research in arthritis applications offers best practices and algorithms for accurate case identification (13). However, not all jurisdictions have access to the same data sources, and the benefit of using more than 1 administrative database for estimating OA incidence and prevalence to increase the capture of cases must be determined. Our study builds and expands on previous research to estimate the incidence and prevalence of OA in Alberta, using administrative health data, and examines how sensitive these estimates are to the case definition for OA, the number of years of historical data, and the data sources accessed.

MATERIALS AND METHODS Data sources. Ethics approval for this study was obtained through the University of Calgary, Conjoint Health Research Ethics Board. We obtained data from 5 Alberta Health administrative databases (descriptions below) to identify individuals with OA who access health care services, paid for by the provincial health care insurance plan. Alberta Health Care Insurance Plan (AHCIP) Population Registry. Individual level demographic data on all

Marshall et al insured persons as of the last day of each fiscal year (March 31), providing population counts. Members of the Armed Forces and the Royal Canadian Mounted Police, federal penitentiary inmates, and Albertans who have opted out of the AHCIP are excluded. All Albertans who are included in the AHCIP have a unique, 9-digit personal health number, which is used when accessing health care services. Discharge Abstract Database (DAD). Admission and inpatient care data for all hospitalized patients, including coded diagnostic, intervention, patient demographic, and administrative information. We obtained data for all patients with OA-related International Classification of Diseases (ICD) codes in any of the 25 DAD diagnosis fields. ICD codes identified as OA-related included those with the first 3 digits 715 or M15 to M19 based on the ninth (9) and tenth (10) revisions of the ICD, respectively (16,17). These ICD codes include OA involving any site, multiple sites, or unspecified sites. ICD-9 code 715 can include OA of the spine, whereas ICD-10 codes M15 to M19 do not include OA of the spine (16,17). Exploratory analysis revealed that providers often omit the fourth and fifth digits following the 715 code, which specify the joint affected. Therefore, we were not able to examine jointspecific OA. Physician Claims Database (claims). Outpatient feefor-service billing data for publicly funded physician services. OA-related visits were identified based on the aforementioned ICD codes in any of the 3 diagnostic fields. Ambulatory Care Classification System (ACCS). Ambulatory visit service utilization data for traditional hospital-based programs (e.g., emergency and day surgery), services delivered in community-based settings (e.g., outpatient clinics), and publicly funded hospital support services such as physiotherapy and occupational therapy. OA-related cases were identified based on the presence of the aforementioned ICD codes in any of the 16 diagnostic fields. Alberta Blue Cross (ABC). Prescription drug use data for Albertans for whom Alberta Health pays ABC premiums (primarily individuals ages $65 years). We used drug identification numbers matched to the anatomical therapeutic chemical codes (developed by the World Health Organization) (18) for musculoskeletal system and analgesics (Table 1). Data were obtained from April 1993 through March 2010, except ACCS data, which were obtained from April 1997 through March 2010. Data were de-identified and assigned a unique patient identifier. An experienced analyst (EE) familiar with Alberta-specific administrative data used the unique patient identifiers to link the 5 data sets. After the 5 data sets were linked, the case definitions (below) were applied to generate the OA patient cohorts. OA case identification. We applied the most current and validated OA case definition for administrative data as our base case (3,8,14,19) based on the following criteria: at least 1 OA-related hospitalization (DAD), or at least 2 OA-related physician visits within 2 years (claims), or at least 2 OA-related ambulatory care visits within 2 years

Using Administrative Health Data to Estimate OA Rates

Table 1.

OA case definitions: base case and alternative scenarios*

Scenario Base case A: Claims, DAD and ACCS, with specific profession and ACCS codes removed B: Claims, DAD and ACCS OA related visits within 5 years C: Claims, DAD and ACCS, with specific profession, service and ACCS codes removed D: Claims only E: Claims, DAD, ACCS, and ABC

F: 1 OA visit in any database

1381

DAD

ACCS

2 visits within 2 years 2 visits within 2 years

1 visit

2 visits within 2 years 2 visits within 2 years

None

2 visits within 5 years 2 visits within 2 years

1 visit

2 visits within 5 years 2 visits within 2 years

None

2 visits within 2 years 2 visits within 2 years 1 visit

1 visit

1 visit

1 visit

2 visits within 2 years

1 visit

1 visit

ABC

Professions and codes removed

Claims†

1. Claims professions‡ 2. ACCS codes§

Any ATC codes#

1. Claims professions‡ 2. Claims service codes¶ 3. ACCS exclusions§ 1. Claims professions‡ 2. Claims service codes¶ 1. Claims professions‡ 2. Claims service codes¶ 3. ACCS exclusions§ None

* OA 5 osteoarthritis; DAD 5 Discharge Abstract Database; ACCS 5 Ambulatory Care Classification System Data; ABC 5 Alberta Blue Cross; ATC 5 anatomical therapeutic chemical. † Excluded for all scenarios: physiotherapists claims and chiropractors claims. ‡ Excluded professions: anesthesiology, anatomical pathology, diagnostic radiology, medical microbiology, and all allied health professionals. § Management Information System codes excluded: undefined code (713508040, 713953000), interventions (714152000), clinical nutrition (714450000), rehabilitation (714500000, 714550000, 714602000, 714604000, 714850000, 714950000, 714955000), and social work (714700000); ACCS codes excluded: interventions (74–76, 78–80, 87), clinical mental health (1058, 1061, 1062), clinical rehabilitation (1101–1106, 1111–1116, 1121–1126, 1131–1134, 1141–1143, 1151–1156), clinical nutrition (1201–1206), clinical social work (1221–1225), clinical psychology (1241–1244), clinical examination/other (2064), and clinical telephone contact (2082). ¶ Claims service codes excluded: “Diagnostic procedures and certain other procedures, for example, dialysis/transfusion, anesthesia, obstetrics, or therapeutic radiation” (2). # ATC codes excluded: M01AB01 indomethacin, M01AB02 sulindac, M01AB05 diclofenac, M01AB08 etodolac, M01AB15 ketorolac, M01AB55 diclofenac in combination, M01AC01 piroxicam, M01AC02 tenoxicam, M01AE01 ibuprofen, M01AE02 naproxen, M01AE03 ketoprofen, M01AE09 flurbiprofen, M01AE11 tiaprofenic, M01AG01 mefanamic acid, M01AH01 celecoxib, M01AX01 nabumetone, N02AA05 oxycodone, N02AD01 pentazocine, N02BA51 codeine in combination, N02BE01 acetaminophen, N02BE51 paracetamol in combination, excluding psycholeptics, and R05DA04 codeine.

(ACCS), and none of the threshold physician or ambulatory care visits being on the same day. These definitions have been used in previous OA administrative data research (2) and have also been validated, demonstrating sensitivity and specificity ranging from 35–52% and 90–95%, respectively (20). Further, research in diabetes mellitus, hypertension, and other chronic diseases have determined the case definition with the highest validity to be 2 physician claims within 2 years, or 1 hospitalization, with a minimum of 3 or more years of data (21–24). We explored alternative case definitions (Table 1) based on previously published research (3,8,14,19) and their impact on OA estimates. OA visits to providers and services that were unlikely to confirm an OA diagnosis (e.g., mammography visits) were excluded based on direct input from experienced clinicians and health services researchers with expertise in administrative data analysis (CB and DM). Incidence and prevalence rates. The incidence rate expresses the number of new cases occurring during a period of time (typically 1 year) (10). Prevalence indicates the number of cases present in a given population at a particular time (10,11). The year of the first visit was assigned as the incident year of a new OA case, providing that the OA case definition criteria were fulfilled (e.g., 1 OA-related hospitalization [DAD] or 2 visits within 2 years for physician claims

and ACCS data sets in the base case). Given that our data began at 1993, we had to assume that the first OA-related visit identified was their first visit. Consequently, there are some patients who are left censored in that they had OA-related visits prior to 1993. Therefore, we were unable to capture individuals as an OA case until their second visit occurred within the 2 year time period (earliest would be 1995), unless it were for a hospitalization where only 1 OA-related hospitalization is required to fulfill the case definition. The OA incidence rate was calculated by dividing the volume of incident OA cases for a given year by the number of person years at risk, which was estimated as the mid-year (September 30) population size less the estimated mid-year number of OA prevalent cases. The OA prevalence rate was calculated as the number of OA cases in the population in the year divided by the mid-year population size. All calculations were performed at the provincial level and age and sex standardized using the direct standardization method with the 2002 AHCIP population. Confidence intervals (CIs) were calculated using the binomial method at the 95% level (25). To gain insight into the contribution of each data set to the identification of OA cases to estimate OA incidence and prevalence, we examined the number of cases arising from each data set that were not previously identified, pro-

1382

Marshall et al

Table 2.

Incidence and prevalence of OA by age group and sex (2008 fiscal year, base case)*

Incidence rate per 1,000 people Age, years All ages ,25 25–29 30–34 35–39 40–44 45–49 50–54 55–59 60–64 65–69 70–74 75–79 80–84 85–89 $90 Overall Crude Standardized (95% CI)

Prevalence rate per 1,000 people

Incidence

Prevalence

Female

Male

Female

Male

Female

Male

Female

Male

8.5 0.5 1.4 2.3 3.6 6.4 10.8 17.4 22.7 27.6 33.2 40.0 48.0 51.8 54.0 55.9

6.5 0.4 1.6 2.1 3.8 5.5 8.4 12.1 16.6 20.2 24.2 30.0 33.0 38.1 43.0 39.0

19,105 287 188 290 453 837 1,537 2,212 2,235 2,119 1,866 1,836 1,915 1,585 1,096 651

13,167 238 210 266 485 722 1,200 1,591 1,685 1,555 1,329 1,283 1,133 827 471 172

102.3 2.7 11.6 17.7 29.3 49.1 80 132.3 201.1 276.6 346.3 417.4 479.9 540.2 583.9 604.9

70.9 2.4 10.2 17.0 27.1 43.7 69.0 99.8 138.9 189.2 251.2 314.1 375.4 440.3 489.2 499.9

177,870 1,551 1,555 2,233 3,688 6,428 11,386 16,817 19,798 21,237 19,459 19,159 19,147 16,525 11,847 7,040

122,995 1,425 1,339 2,160 3,465 5,729 9,863 13,121 14,100 14,566 13,793 13,429 12,889 9,559 5,356 2,201

7.5 8.6 (8.3–8.9)

32,272

86.6 80.3 (80.0–80.5)

300,865

* OA 5 osteoarthritis; 95% CI 5 95% confidence interval.

gressing from claims to DAD to ACCS data. This sequence was chosen as all OA patients diagnosed by a physician would be recorded in the claims data, while only a subset would be hospitalized due to OA or see a publicly funded allied health professional, and therefore appear in the DAD or ACCS data sets. Due to the case definition criteria with respect to the intervisit period, 2008 is the most recent year for which incidence or prevalence could be reported for most scenarios, as any “first” visits that occur in 2009 or 2010 may have a “second” visit beyond the scope of the data set. Similarly, 2005 is the most recent year that can be reported for the 5-year criteria (Scenario B).

RESULTS OA incidence rate and prevalence. Alberta’s crude and age- and sex-standardized incidence rates for 2008 were estimated to be 7.5 cases per 1,000 population and 8.6 per 1,000, respectively (Table 2). Overall, women had higher OA incidence rates than men (8.5 cases versus 6.5 cases per 1,000 population). Rates increased with age for both sexes (Table 2). Applying these incidence rates to the entire population of Alberta yields an estimate of more than 32,000 new OA cases in Alberta in fiscal year 2007–2008. The overall crude prevalence for 2008 was estimated to be 86.6 per 1,000 population and 80.3 per 1,000 population when age and sex standardized (Table 2) in the base case. The prevalence was higher in women in almost all age groups and increases dramatically after age 50 years, peaking at approximately 600 per 1,000 in women in the

$90 years age group (Table 2). The largest number of Albertans experiencing OA are ages 60–64 years (35,803). Effect of multiple data sets. The claims data alone identified the majority (80–90%) of OA incident cases each year (Table 3). The DAD and ACCS data increased the number of incident cases by 5–10% and 2–10%, respectively, for each year over the time period. The ABC data had a negligible impact (see Supplementary Table 1, available in the online version of this article at http:// onlinelibrary.wiley.com/doi/10.1002/acr.22612/abstract). Effect of years of data. In our analysis, prevalence is cumulative across the years of data; therefore, the estimates reflect the number of years of data available. The prevalence starts to plateau as we approach 15 years of longitudinal data, by which time most prevalent cases in the population are likely identified (Figure 1). The incidence rates stabilize within a few years and decrease gradually in the later years (for associated CIs, see Supplementary Table 2, available in the online version of this article at http:// onlinelibrary.wiley.com/doi/10.1002/acr.22612/abstract). Sensitivity analyses of alternative case definitions. Comparing the estimates of incidence and prevalence from the sensitivity analysis demonstrates the impact of altering the case definition (Figure 2). In all scenarios, the incidence and prevalence estimates are significantly different from the base case at the 95% level. The incidence rate in scenario F (1 OA-related visit in any database) (16.9 per 1,000 population; 95% CI 16.7–17.1) is nearly double that of the base case (8.6 per 1,000; 95% CI 8.5–8.7),

Using Administrative Health Data to Estimate OA Rates

1383

Table 3. Relative contribution of administrative databases to identify incident OA cases (base case)* OA cases, no. (% of total)† Year 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Claims 19,672 19,697 19,483 22,163 21,249 20,168 20,261 21,386 21,797 21,127 20,611 20,810

(89.1) (87.6) (88.0) (88.7) (80.7) (80.5) (83.0) (84.4) (87.4) (87.4) (88.3) (88.4)

DAD

ACCS‡

Total, no.

2,413 (10.9) 2,214 (10.0) 2,057 (9.3) 2,152 (8.6) 2,394 (9.1) 2,463 (9.8) 1,962 (8.0) 1,711 (6.8) 1,511 (6.1) 1,380 (5.7) 1,168 (5.0) 1,228 (5.2)

0 (0.0) 546 (2.4) 594 (2.7) 678 (2.7) 2,704 (10.3) 2,436 (9.7) 2,196 (9.0) 2,250 (8.9) 1,641 (6.6) 1,674 (6.9) 1,570 (6.7) 1,496 (6.4)

22,085 22,457 22,134 24,993 26,347 25,067 24,419 25,347 24,955 24,181 23,349 23,534

* OA 5 osteoarthritis; DAD 5 Discharge Abstract Database; ACCS 5 Ambulatory Care Classification System Data. † Results reflect only the insured population of Alberta. ‡ ACCS data were only collected in Alberta as of April 1, 1997 and the scope of collection changed over time.

and prevalence (144.5 per 1,000 population; 95% CI 144.1– 144.9) increases significantly from the base case (80.3 per 1,000 population; 95% CI 80.0–80.5) by 223,030 cases in Alberta in 2008. Requiring only 1 indication of OA, scenario F captures all recorded instances of confirmed or suspected OA and produces the highest estimates. As expected, removing specific provider visits and services unlikely to be OA related lowered incidence and prevalence estimates. Among these, claims data alone, with specific provider visits and services removed (Scenario D, physician claims only), produced the lowest estimates. Comparing incidence and prevalence of Scenario E, which included ABC data, and Scenario C, which did not include ABC data but was otherwise identical, indicates that the prescription drug data had no significant impact on the estimates.

Figure 1. Effect of number of years of historical data on osteoarthritis incidence and prevalence rate estimation; 95% confidence intervals can be seen in Supplementary Table 2, available in the online version of this article at http://onlinelibrary.wiley. com/doi/10.1002/acr.22612/abstract.

Figure 2. Impact of osteoarthritis (OA) case definition on ageand sex-standardized (A) incidence of OA and (B) prevalence of OA in 2008. * 5 scenario A: claims, Discharge Abstract Database (DAD), and Ambulatory Care Classification System data (ACCS), with specific profession and ACCS codes removed; † 5 scenario B: claims, DAD, and ACCS OA-related visits within 5 years; ‡‡ 5 data shown for 2005 due to need for 5-year time period; ‡ 5 scenario C: claims, DAD, and ACCS, with specific profession, service, and ACCS codes removed; § 5 scenario D: claims only; ** 5 scenario E: claims, DAD, ACCS, and Alberta Blue Cross; †† 5 scenario F: 1 OA visit in any database.

Increasing the intervisit time horizon from 2 to 5 years (base case versus Scenario B) captured more cases but also increased the amount of historic data required and resulted in a statistically significant increase in incidence (to 10.4 from 8.6 per 1,000 population) and in prevalence (to 82.6 from 80.3 per 1,000 population). Comparing Scenarios A and C, which are identical apart from the removal of specific provider visits and service codes in the latter, demonstrates little impact, while omitting both specific professions within the claims data and selected ACCS codes in Scenario A decreased rates significantly compared to the base case. Comparing scenarios A and D (physician claims only) indicates that including only physician claims data yielded significantly lower estimates (7.7 per 1,000 population; 95% CI 7.6–7.8) than scenario A (8.2 per 1,000 population; 95% CI 8.1–8.3).

DISCUSSION Administrative health databases are increasingly being used for epidemiologic and outcomes research, health care quality measurement and management, and health services

1384 population-based research (15,20,26,27). We used administrative data to estimate OA incidence rates and prevalence for Alberta. The base case age- and sex-standardized incidence and prevalence rate estimates were 8.6 and 80.3 cases per 1,000 population, respectively. Both rates rose substantially over age 50 years and were higher for women. More than 10% of Canadians are impacted by OA (11). The World Health Organization estimates that worldwide, 9.6% of men and 18.0% of women ages $60 years have symptoms of OA (28). Hip/knee OA is the most common form of the disease, and based on data from the Global Burden of Disease 2010 study, it is estimated that hip/ knee OA is the 11th highest contributor to disability internationally (29). Our estimate of OA prevalence within the province of Alberta (8.03% in 2008) was lower than these global estimates. This was likely because some of the other estimates are self-reported, which tend to be higher, and OA tends to be underreported in administrative data, as reflected by the sensitivity ranging between 35% and 52% in validation studies (20). Overall, our estimates are comparable but lower than some previous findings based on administrative data. Sun et al found the incidence rate among Albertans to be 10.4 per 1,000 population in 2002 (30). Similarly, Kopec et al estimated OA incidence and prevalence rates to be 11.7 and 107.8 per 1,000 population, respectively, in British Columbia in 2001 (2). However, both of these studies defined OA cases based on only claims data and 1 physician visit rather than the 2 visits within 2 years applied in our base case scenario. Therefore, these cases are more comparable to Scenario F (1 OA-related visit in any database over the range of years for which we have data; April 1993 to March 2010), for which we estimated incidence at 16.9 cases per 1,000 population. Our estimates are likely higher because Scenario F included the DAD and ACCS data sources. Applying criteria similar to our base case, Kopec et al obtained prevalence estimates of 69 and 48 per 1,000 for females and males respectively, and incidence estimates of 7.7 and 5.5 per 1,000 for females and males respectively (2). These prevalence rates are lower than our estimates presented in Table 2. This difference is likely because they used fewer years of historic data; their results indicate that the rates had not yet plateaued, and additional years may have increased their prevalence results. Further, they used only 2 data sources equivalent to our DAD and claims data. Omitting ACCS data contributed 6–7% of the cases we identified, and they only included individuals ages .20 years, while we included all ages. The Arthritis Alliance of Canada estimated the prevalence of OA to be 13%, or 130 per 1,000 population, based on applying age- and sex-specific prevalence determined by Kopec et al (2,3) to national population data (11). Other published OA prevalence estimates are higher than ours due to the use of self-report data encompassing all types of arthritis (12,31). Self-reported data are known to overestimate prevalence (7) and often do not report OA-specific findings (2). Our comparison of the impact of altering the case definition on OA incidence and prevalence estimates are of interest since not all jurisdictions have ready access to the

Marshall et al same data sources or volume of historic data. We explored this further by evaluating the effect each data source and the number of years of data analyzed had on the outcomes for each scenario (Supplementary Table 1, available in the online version of this article at http://onlinelibrary.wiley. com/doi/10.1002/acr.22612/abstract). Relying solely on claims data would have missed 12% of OA cases, therefore linking data sources was found to be beneficial as these remaining cases were identified almost equally among the DAD and ACCS data sets. Although relying on claims data alone may underestimate the rates, it may be sufficient for monitoring health resource use even if post hoc adjustments are necessary. We also found that adding prescription drug data (Scenario E) had a negligible impact on incidence and prevalence estimates. Scenario E estimates of incidence (8.2 per 1,000 population; 95% CI 8.1–8.3) and prevalence (77.2 per 1,000 population; 95% CI 76.9–77.5) were not found to be significantly different than Scenario C (incidence 8.3 per 1,000 population; 95% CI 8.2–8.4 and prevalence 76.8 per 1,000 population; 95% CI 76.6–77.1). Furthermore, given the variation, scope, availability of prescription drug data, and the nonspecificity of many OA drugs, a case definition that does not require these data may be advantageous for ongoing monitoring provincially and nationally. Long-term longitudinal data are critical for estimating prevalence in chronic diseases such as OA (2). Our results indicate that OA prevalence rate estimates required approximately 15 years of longitudinal data to plateau. Estimates based on fewer years will likely underestimate OA prevalence. When estimating incidence, some changes are expected due to changes in risk factors and individual behavior, and the effect of institutionalization of older people with OA over time (7–10). Additionally, there are individuals in the population with OA who may experience symptoms for many years prior to obtaining a diagnosis or seeking care (32). Our prevalence estimates capture individuals who presumably have persistent or recurrent symptoms of OA (such that they have at least 2 visits in 2 years) and therefore may underestimate the true prevalence of OA. Our study used administrative data from the province of Alberta. Although individuals responsible for ICD coding in the DAD receive standardized training in Canada (33), there may still be coding variations across provinces that could limit the generalizability of these findings. Previous studies have compared the validity of case identification using administrative data against chart review in the areas of hypertension and diabetes mellitus, and found that the sensitivity results (diabetes mellitus range 90.2–92.1%, hypertension range 74.0–76.0%) are not significantly different between 2 provinces (Alberta and British Columbia) (21,24). Administrative data sources have many advantages, such as ease of access and study, the wide range of chronic diseases and comorbidities that are captured, and the ability to provide both cross-sectional and longitudinal data (19). Longitudinal administrative data can be used to inform simulation models, such as the system dynamics model developed by Vanderby et al specifically for OA (34). These models can be used for understanding the

Using Administrative Health Data to Estimate OA Rates structure and behavior of complex health systems over time (35). The authors have used the model to create longitudinal cohorts of patients to determine how long patients with OA spend in different stages of care across the continuum of care, estimate health care resource use, revision rates for hip or knee arthroplasty, as well as determine the impact of other changes to health care service delivery such as wait time management strategies (34). There are limitations to our results. ACCS data prior to 1997 were not available. Therefore, based on our sensitivity analysis that estimated the contributions of ACCS data to OA case identification, approximately 6% of OA cases may have been missed in the pre-1997 analyses. This may result in an underestimation of OA prevalence. The diagnostic codes submitted by physicians in the claims data may not be clinically accurate; therefore, OA treatments may not be reflected in the codes reported. Additionally, OA codes may be recorded when physicians are querying an OA diagnosis. These 2 issues may lead to under- and overestimates of OA cases, respectively; our base case definition criteria were designed to mitigate these issues, yet the results may still be affected to an unknown extent. The sensitivity and specificity of OA case definition are not high; however, these are the standard algorithms accepted for administrative database OA research. Further, previous research in other chronic disease areas have determined the case definition with the highest validity to be 2 physician claims within 2 years, or 1 hospitalization, with a minimum of 3 or more years of data (21–24). Our estimates include only individuals who have accessed health care services paid for by the provincial health care insurance plan. Therefore, individuals with OA symptoms who have not accessed health care services for their OA (32) or who have accessed only private health care services would not be included in the OA cohort. Further, additional factors that may influence OA incidence rates, such as obesity, are not reliably reflected in the data. For example, patient height and weight information are not routinely collected; the resulting inability to account for obesity rates may also contribute variations in reported results as changes in these rates may contribute to changes in OA incidence. It is important to have a sound basis to estimate the future burden of chronic diseases and their associated health services use, particularly as the population ages, to plan for appropriate health service needs (26). We have estimated OA incidence and prevalence in Alberta and demonstrated that these estimates are mostly defined through claims data, require a long period of longitudinal data, and are sensitive to alternative case definitions.

AUTHOR CONTRIBUTIONS All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Marshall had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study conception and design. Marshall, Vanderby, Barnabe, Maxwell, Mosher. Acquisition of data. Marshall, Frank.

1385 Analysis and interpretation of data. Marshall, Vanderby, Barnabe, MacDonald, Maxwell, Mosher, Wasylak, Lix, Enns, Frank, Noseworthy.

REFERENCES 1. Finlayson G, Ekuma O, Yogendran M, Burland E, Forget E. The additional cost of chronic disease in Manitoba. Winnipeg (Canada): Manitoba Centre for Health Policy; 2010. 2. Kopec JA, Rahman M, Berthelot J, Le Petit C, Aghajanian J, Sayre EC, et al. Descriptive epidemiology of osteoarthritis in British Columbia, Canada. J Rheumatol 2007;34:386–93. 3. Kopec JA, Rahman MM, Sayre EC, Cibere J, Flanagan WM, Aghajanian J, et al. Trends in physician-diagnosed osteoarthritis incidence in an administrative database in British Columbia, Canada: 1996–1997 through 2003–2004. Arthritis Rheum 2008;59:929–34. 4. Kopec JA, Sayre EC, Flanagan WM, Fines P, Cibere J, Rahman MM, et al. Development of a population-based microsimulation model of osteoarthritis in Canada. Osteoarthritis Cartilage 2010;18:303–11. 5. Srikanth V, Fryer J, Zhai G, Winzenberg T, Hosmer D, Jones G. A meta-analysis of sex differences prevalence, incidence and severity of osteoarthritis. Osteoarthritis Cartilage 2005;13:769–81. 6. Anderson SA, Loeser RF. Why is osteoarthritis an age-related disease? Best Pract Res Clin Rheumatol 2010;24:15–26. 7. Woolf A, Pfleger B. Burden of major musculoskeletal conditions. Bull World Health Organ 2003;81:646–56. 8. Felson DT, Lawrence RC, Dieppe PA, Hirsch R, Helmick CG, Jordan JM, et al. Osteoarthritis: new insights. Part 1: the disease and its risk factors. Ann Int Med 2000;133:635–46. 9. Zhang Y, Jordan JM. Epidemiology of osteoarthritis. Clin Geriatr Med 2010;26:355–69. 10. Losina E, Walensky RP, Reichmann WM, Holt HL, Gerlovin H, Solomon DH, et al. Impact of obesity and knee osteoarthritis on morbidity and mortality in older Americans. Ann Int Med 2011;154:217–26. 11. Bombardier C, Hawker G, Mosher D. The impact of arthritis in Canada: today and over 30 years: Arthritis Alliance of Canada; 2011. URL: http://www.arthritisalliance.ca/images/ PDF/eng/Initiatives/20111022_2200_impact_of_arthritis.pdf. 12. Statistics Canada. Summary tables: arthritis, by sex, and by province and territory. 2014. URL: http://www.statcan.gc. ca/tables-tableaux/sum-som/l01/cst01/health52a-eng.htm. 13. CANRAD Network. Advancing the science of administrative health data research and surveillance for improving the health of people with arthritis: about CANRAD. 2013. URL: http://www.canradnetwork.ca/about-canrad/. 14. Widdifield J, Labrecque J, Lix L, Paterson JM, Bernatsky S, Tu K, et al. Systematic review and critical appraisal of validation studies to identify rheumatic diseases in health administrative databases. Arthritis Care Res (Hoboken) 2013;65:1490–503. 15. Goldfield N, Willani J. The use of administrative data as the first step in the continuous quality improvement process. Am J Med Qual 1996;11:S35–8. 16. World Health Organization. International Classifications of Disease (ICD). 2015. URL: http://www.who.int/classifications/ icd/en/. 17. Centers for Disease Control and Prevention. International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). 2013. URL: http://www.cdc.gov/nchs/icd/ icd9cm.htm. 18. World Health Organization Centre for Drug Statistics Methodology. Guidelines for ATC classification and DDD assignment. 2013. URL: http://www.whocc.no/filearchive/publications/ 1_2013guidelines.pdf. 19. Lix L, Yogendran M, Burchill C, Metge C, McKeen N, Moore D, et al. Defining and validating chronic diseases: an administrative data approach. 2006. URL: http://mchp-appserv. cpe.umanitoba.ca/reference/chronic.disease.pdf. 20. Lix L, Yogendran M, Mann J. Defining and validating chronic diseases: an administrative data approach. An update

1386

21. 22. 23. 24. 25.

26. 27. 28.

with ICD-10-CA. Winnipeg (Manitoba, Canada): Manitoba Centre for Health Policy; 2008. Chen G, Khan N, Walker R, Quan H. Validating ICD coding algorithms for diabetes mellitus from administrative data. Diabetes Res Clin Pract 2010;89:189–95. Gershon A, Wang C, Guan J, Vasilevska-Ristovska J, Cicutto L, To T. Identifying patients with physician-diagnosed asthma in health administrative databases. Can Respir J 2009;16:183–8. Tu K, Campbell NR, Chen Z, Cauch-Dudek K, McAllister FA. Accuracy of administrative databases in identifying patients with hypertension. Open Med 2007;1:E18–E26. Quan H, Khan N, Hemmelgarn BR, Tu K, Chen G, Campbell N, et al. Validation of a case definition to define hypertension using administrative data. Hypertension 2009;54:1423–8. Boyle P, Parkin D. Statistical methods for registries. In: Jensen O, Parkin D, MacLennan R, Muir C, Skeet R, editors. Cancer registration: principles and methods. Lyon (France): International Agency for Research on Cancer; 1991. Harrold L, Yood R, Andrade S, Reed J, Cernieux J, Straus W, et al. Evaluating the predictive value of osteoarthritis diagnoses in an administrative database. Arthritis Rheum 2000;43:1881–5. Schneeweiss S, Avorn J. A review of uses of health care utilization databases for epidemiologic research on therapeutics. J Clin Epidemiol 2005;58:323–37. World Health Organization. Chronic rheumatic conditions. 2015. URL: http://www.who.int/chp/topics/rheumatic/en/.

Marshall et al 29. Cross M, Smith E, Hoy D, Nolte S, Ackerman I, Fransen M, et al. The global burden of hip and knee osteoarthritis: estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis 2014;73:1323–30. 30. Sun J, Gooch K, Svenson L, Bell N, Frank C. Estimating osteoarthritis incidence from population-based administrative health care databases. Ann Epidemiol 2007;17:51–6. 31. Public Health Agency of Canada. Life with arthritis in Canada: a personal and public health challenge. Ottawa (Ontario, Canada): Public Health Agency of Canada; 2010. 32. MacDonald KV, Sanmartin C, Langlois K, Marshall DA. Symptom onset, diagnosis and management of osteoarthritis. Health Rep 2014;25:10–7. 33. Reid AY, St. Germaine-Smith C, Liu M, Sadiq S, Quan H, Wiebe S, et al. Development and validation of a case definition for epilepsy for use with administrative health data. Epilepsy Res 2012;102:173–9. 34. Vanderby SA, Carter MW, Noseworthy T, Marshall DA. Modelling the complete continuum of care using system dynamics: the case of osteoarthritis in Alberta. J Simulat 2015. E-pub ahead of print. 35. Marshall DA, Burgos-Liz L, Ijzerman MJ, Osgood ND, Padula WV, Higashi MK, et al. Applying dynamic simulation modeling methods in health care delivery research: the SIMULATE checklist. Report of the ISPOR Simulation Modeling Emerging Good Practices Task Force. Value Health 2015;18:5–16.