Validation of International Classification of Diseases Code–Based Case Definitions of Immune Checkpoint Inhibitor–Associated Inflammatory Arthritis From Administrative Health Data

Abstract

Objective

Immune checkpoint inhibitors (ICIs) for cancer can lead to immune‐related adverse events, including ICI‐associated inflammatory arthritis (ICI‐IA). There are no validated International Classification of Diseases (ICD) code–based case definitions for ICI‐IA.

Methods

We conducted a validation study using the Canadian Research Group of Rheumatology in Immuno‐Oncology (CanRIO) database. Rheumatologist‐confirmed ICI‐IA was the reference standard, defined as new onset of at least one swollen joint on examination or synovitis on imaging following ICI exposure, without another cause or pre‐existing IA. We compared ICD‐9 (714.x and 696.0) and ICD‐10 (M05.x, M06.x, and M07.x) IA diagnostic codes against the reference standard. Seven core case definitions of different combinations of ICD codes from the Physician Claims Database (PCD) and the Discharge Abstract Database were tested. Results were stratified by sex. Sensitivity testing with additional criteria was also evaluated.

Results

We included 228 patients in the final analysis: 100 with ICI‐IA and 128 without ICI‐IA. Sensitivity of the tested case definitions ranged from 1.0% to 88.0%, whereas specificity ranged from 86.7% to 100.0%. The case definition with at least one PCD IA code achieved the best balance of sensitivity (88.0%, 95% confidence interval [CI] 81.6%–94.4%) and specificity (86.7%, 95% CI 80.8%–92.6%). Case definition performances were similar between sexes. Additional criteria minimally improved specificity but sacrificed sensitivity.

Conclusion

ICD code–based case definitions of ICI‐IA can accurately detect ICI‐IA and can be used to support ICI‐IA surveillance and research with administrative health data.

INTRODUCTION

Immune checkpoint inhibitors (ICIs) are now widely used to treat numerous malignancies, with immunotherapy representing the fourth pillar of cancer therapies alongside chemotherapy, radiation, and surgery. ¹ ICIs function by blocking immune regulatory checkpoints, such as programmed death 1 (PD‐1); its ligand, programmed death ligand 1 [PD‐L1]; and cytotoxic T‐lymphocyte–associated antigen 4 (CTLA‐4), thereby enhancing the immune system to attack tumor cells and ultimately prolonging survival in many patients. ² Although ICIs offer significant clinical benefits, they can cause off‐target effects known as immune‐related adverse events (irAEs), which may affect multiple organs. ³

Among these, rheumatic irAEs have been increasingly recognized in small case series and observational studies, though they remain poorly described in randomized controlled trials. ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ ICI‐associated inflammatory arthritis (ICI‐IA), the most common rheumatic irAE, is estimated to occur in 1% to 11% of ICI users. ¹¹ , ¹² , ¹³ , ¹⁴ However, the true incidence is unknown as most of the reported data on ICI‐IA come from clinical case series, and ICI clinical trials tend to report symptoms, such as arthralgias and myalgias, and not the underlying cause of these symptoms (ie, osteoarthritis vs IA vs soft tissue disorders vs metastatic cancer). ¹¹ , ¹⁴ , ¹⁵ Additionally, the reporting of irAE in ICI clinical trials is generally completed by the oncology team, and not rheumatologists, making accurate incidence rates for ICI‐IA difficult to ascertain.

Although studies have been done for detection of irAEs in general using administrative health data and electronic health records, ¹⁶ , ¹⁷ International Classification of Diseases (ICD) code–based case definitions for rheumatic irAEs have not been validated, an important standard for use of ICD code–based case definitions in administrative database research. ¹⁸ Despite this, several administrative database studies have been published examining the frequency and predictors and outcomes of rheumatic irAEs, including ICI‐IA, making the validity of these studies difficult to assess. ¹³ , ¹⁹ , ²⁰ , ²¹ This underscores the urgent need to validate ICD code–based case definitions for rheumatic irAEs.

One of the challenges of validating administrative case definitions for rheumatic irAEs is that it requires a relatively large cohort of confirmed cases and confirmed noncases. The Canadian Research Group of Rheumatology in Immuno‐Oncology (CanRIO) database is a national database of ICI users referred to rheumatology for assessment of possible rheumatic irAEs, which includes rheumatologist‐confirmed cases (and noncases) of ICI‐IA. ⁸ By leveraging the CanRIO database, our study aims to validate ICD code–based case definitions of the most common rheumatic irAE: ICI‐IA.

METHODS

Data sources

We conducted a validation study analyzing data from the Alberta cohort of the CanRIO database (CanRIO‐AB). Alberta is the fourth most populous province in Canada, with an estimated population of 4.9 million people in 2024. ²² Every resident of Alberta is assigned a unique personal health number that provides universal access to all health services in the province and can be used to track individual health care use across settings (outpatient clinics, emergency departments, hospitalizations, laboratories, diagnostic imaging, pharmacies). These identifiers were used to link individuals in CanRIO‐AB to their Alberta provincial health administrative records.

Baseline patient characteristics including demographics and cancer‐ and ICI‐related data were ascertained from CanRIO‐AB. IA ICD codes were ascertained from the Physician Claims Database (PCD) and the Discharge Abstract Database (DAD) from three years before ICI initiation up to July 31, 2024, death, or loss of registration, whichever came sooner. The PCD employs ICD, Ninth Revision, Clinical Modification (ICD‐9‐CM) codes, whereas the DAD (which captures all hospitalizations) uses ICD, Tenth Revision, Canadian Enhancements (ICD‐10‐CA) codes. Deaths were ascertained from Vital Statistics, which captures all births and deaths in Alberta. The study was approved by the Health Research Ethics Board for the University of Alberta (Pro00118986), which granted a waiver for individual consent as investigators were only supplied with data after de‐identification. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for observational cohort studies. ²³ Patients or the public were not involved in the planning or conduct of this research study.

CanRIO database

CanRIO is a network of rheumatologists from 11 institutions across Canada specializing in rheumatic irAEs. The retrospective CanRIO database was established in January 2019 and includes patients assessed between January 2013 to May 30, 2024. ⁸ , ²⁴ It includes all patients treated with ICIs referred to CanRIO rheumatologists for assessment of possible rheumatic irAEs or pre‐existing rheumatic diseases. Chart review was conducted for each patient, and pertinent data, including patient demographics, history of pre‐existing rheumatic diseases, oncologic history, irAE manifestations, treatments, and outcomes, were entered into the CanRIO database. All patients in CanRIO‐AB with an ICI initiation date and final diagnosis were included. Patients who could not be linked to the provincial administrative databases were excluded.

ICI‐IA reference cases

The reference standard for ICI‐IA was rheumatologist‐confirmed ICI‐IA in the CanRIO database, defined as the new onset of at least one swollen joint on examination or synovitis on imaging following ICI exposure, without another cause or pre‐existing IA. All other cases in CanRIO‐AB (ie, patients referred to rheumatologists for assessment of rheumatologic symptoms but not confirmed by the rheumatologist to have met criteria for ICI‐IA, including those with existing IA before ICI exposure) were classified as not having ICI‐IA and considered noncases.

ICI‐IA administrative data case definitions

There are no ICI‐IA specific ICD codes. We selected ICD‐9 and ICD‐10 codes that represented IA. ICD‐9 codes included rheumatoid arthritis and other IA (714.x) and psoriatic arthritis (696.0). ICD‐10 codes included seropositive rheumatoid arthritis (M05.x), seronegative rheumatoid arthritis and other IA (M06.x), and psoriatic and inflammatory bowel disease–associated arthritis (M07.x).

IA ICD codes were ascertained three years before ICI initiation and any time post ICI initiation. In order to discriminate flares of pre‐existing IA from ICI‐IA, an individual with an IA ICD code within three years before ICI initiation was classified as negative for ICI‐IA, regardless of the presence of any IA codes ascertained after ICI initiation. We tested seven core case definitions that included different combinations of IA codes from PCD and DAD (Table 1). For our core set of case definitions, the first instance of IA code could occur any time after ICI initiation, but for all case definitions that required a minimum of at least two codes, the codes had to be within three years of the first code. Multiple records of IA codes registered for the same day for the same setting (PCD or DAD) were considered a single instance of IA code in that setting.

Table 1.

Case definitions^*

Case definition	Abbreviation
≥1 Physician Claims Database code ICD‐9 714.x or 696.0	≥1P
≥2 Physician Claims Database codes ICD‐9 714.x or 696.0	≥2P
≥3 Physician Claims Database codes ICD‐9 714.x or 696.0	≥3P
≥1 Discharge Abstract Database code ICD‐10 M05.x, M06.x, or M07.x	≥1H
1 Discharge Abstract Database code ICD‐10 M05.x, M06.x, or M07.x or 1 Physician Claims Database codes ICD‐9 714.x or 696.0	≥1H or ≥1P
1 Discharge Abstract Database ICD‐10 M05.x, M06.x, or M07.x or 2 Physician Claims Database codes ICD‐9 714.x or 696.0	≥1H or ≥2P
1 Discharge Abstract Database code ICD‐10 M05.x, M06.x, or M07.x or 3 Physician Claims Database codes ICD‐9 714.x or 696.0	1H or ≥3P

^*ICD, International Classification of Diseases; P, Physician Claims Database code, H, Discharge Abstract Database code.

Sensitivity analysis

Nine sets of sensitivity analyses were conducted, each with additional criteria added to the core algorithms. (1) IA that develops years after ICI discontinuation may not be related to ICI exposure. Thus, we conducted a sensitivity analysis in which we reclassified reference ICI‐IA cases with onset of symptoms beyond one year (365 days) after ICI discontinuation as not ICI‐IA (ICI‐IA n = 96, no ICI‐IA n = 132), and correspondingly, first IA codes had to occur within the year of ICI discontinuation. (2) Because the vast majority of IA codes were rheumatoid arthritis codes (714.x, M05.x, and M06.x), we evaluated case definitions using only rheumatoid arthritis codes. (3) We restricted case definitions to IA codes from rheumatologists only or (4) to IA codes from internal medicine physicians and rheumatologists only. (5) We examined the addition of IA codes from the National Ambulatory Care Reporting System (NACRS). NACRS captures all emergency department visits using ICD‐10‐CA codes, which were treated as either PCD or DAD codes. Although Alberta has NACRS data for all institutions in the province, we did not include NACRS codes in our core case definition algorithms for this study because many other jurisdictions do not have NACRS data. (6) Case definitions requiring a minimum of at least two codes were restricted to multiple codes having to be within one year of the first code or (7) within two years of the first code. (8) We examined the addition of polymyalgia rheumatica (PMR) codes (ICD‐9 725, ICD‐10 M35.3) to the core algorithms. (9) We examined the addition of at least one filled prescription of a glucocorticoid or disease‐modifying antirheumatic drug (DMARD) within 40 days (before or after) of the first code of ICI‐IA (Supplementary Table 1).

Statistical analysis

Summary statistics were used to describe the baseline characteristics of the CanRIO‐AB cohort, stratified by rheumatologist‐confirmed ICI‐IA status. For each of the seven core case definitions and for each of the sensitivity analyses, we estimated sensitivity (true positives/[true positives + false negatives]), specificity (true negatives/[true negatives + false positives]), positive predictive value (PPV = true positives/[true positives + false positives]), and negative predictive value (NPV = true negatives/[true negatives + false negatives]), along with their corresponding 95% confidence intervals (CIs) and C‐statistic. For the seven core case definitions, with a sample size of 228 cases, including 100 ICI‐IA cases, we could detect 0.9 sensitivity and 0.9 specificity (lower bound of 0.8), with power of 0.8 and alpha of 0.05. ²⁵ Sex‐stratified analysis was performed to test for sex differences in core case definitions. All analyses were conducted using SAS 9.4 (SAS Institute Inc).

Data availability statement

This study is based on data provided by Alberta Health and Alberta Health Services. We are not able to make our data set available to other researchers due to our contractual data disclosure agreement with Alberta Health Services, which is the data custodian, and the restrictions of the Alberta Health Information Act. Inquiries regarding access to the data can be made to health.resdata@gov.ab.ca.

RESULTS

There were a total of 233 individuals in the CanRIO‐AB database as of May 30, 2024, all of whom had a final rheumatologist‐assigned diagnosis (ICI‐IA vs other diagnoses), but four did not have an ICI start date, and one did not have a provincial health registry record. The final study cohort included 228 individuals: 100 in the rheumatologist‐confirmed ICI‐IA group and 128 in the no ICI‐IA group (control group; Figure 1).

Figure 1.

Cohort selection. CanRIO‐AB, Canadian Research Group of Rheumatology in Immuno‐Oncology Alberta; ICI, immune checkpoint inhibitor; ICI‐IA, immune checkpoint inhibitor–associated inflammatory arthritis.

The median age was similar between the two groups: 71.0 years (interquartile range [IQR] 63.0–77.0) in those with ICI‐IA and 70.0 years (IQR 61.0–76.7) in those without (Table 2). A slightly higher proportion of men was observed in the ICI‐IA group (61.0%) compared to the control group (53.1%). Smoking status and alcohol consumption were similar between the two groups. A higher percentage of patients in the control group had pre‐existing autoimmune disease, as compared with the ICI‐IA group (29.7% vs 16.0%), with 15.6% of the control group having had pre‐existing IA and, by definition, none of the ICI‐IA group having had pre‐existing IA.

Table 2.

Baseline characteristics of the participants^*

Variable	CanRIO‐AB
	N = 228
	ICI‐IA	No ICI‐IA
	n = 100	n = 128
Age, median (IQR)	71.0 (63.0–77.0)	70.0 (61.0–76.7)
Sex, n (%)
Women	39 (39.0)	60 (46.9)
Men	61 (61.0)	68 (53.1)
Smoking status, n (%)
Smoked in the past	49 (49.0)	60 (46.9)
Never smoked	39 (39.0)	49 (38.3)
Current smoker	12 (12.0)	19 (14.8)
Alcohol consumption, n (%)
<1 drink/week	71 (71.0)	106 (82.8)
1–7 drinks/week	26 (26.0)	18 (14.1)
>7 drinks/week	3 (3.0)	4 (3.1)
Pre‐existing autoimmune diseases, n (%)	16 (16.0)	38 (29.7)
Pre‐existing IA, n (%)	0 (0.0) a	20 (15.6)
Tumor group, n (%)
Skin	39 (39.0)	42 (32.8)
Lung & bronchus	36 (36.0)	47 (36.7)
Genitourinary	15 (15.0)	17 (13.3)
Gastrointestinal	3 (3.0)	3 (2.3)
Gynecological	4 (4.0)	8 (6.2)
Endocrine & neuroendocrine	0 (0.0)	1 (0.8)
Primary unknown/other	4 (4.0)	10 (7.8)
ICI therapy, n (%)
PD‐1/PD‐L1 inhibitor monotherapy	76 (76.0)	90 (70.3)
CTLA‐4 inhibitor monotherapy	0 (0.0)	0 (0.0)
Combination ICI therapy	20 (20.0)	29 (22.7)
Unknown b	4 (4.0)	9 (7.0)
Concurrent chemotherapy, n (%)	9 (9.0)	23 (18.0)
Cancer stage, n (%)
Stage I	0 (0.0)	1 (0.8)
Stage II	3 (3.0)	11 (8.6)
Stage III	34 (34.0)	39 (30.5)
Stage IV	63 (63.0)	77 (60.2)
Nonrheumatologic irAEs, n (%)	32 (32.0)	41 (32.0)
CanRIO‐AB observation time from ICI initiation in months, median (IQR)	33.9 (20.2–46.9)	24.0 (12.3–50.3)
Provincial registry observation time from ICI initiation in months, median (IQR)	34.4 (20.1–55.2)	23.5 (12.0–50.9)
Death, n (%)	39 (39.0)	49 (38.3%)

^*CanRIO‐AB, Canadian Research Group of Rheumatology in Immuno‐Oncology‐Alberta Cohort; CTLA‐4, cytotoxic T‐lymphocyte–associated antigen 4; IA, inflammatory arthritis; ICI, immune checkpoint inhibitors; ICI‐IA, immune checkpoint inhibitor–associated inflammatory arthritis; IQR, interquartile range; irAE, immune‐related adverse events; PD‐1, programmed death 1; PD‐L1, programmed death ligand 1.

^a Patients with pre‐existing IA by definition could not have ICI‐IA.

^b Blinded clinical trial.

Skin cancer and lung and bronchus cancer were the most common tumor groups, with the breakdown by tumor group appearing similarly in both ICI‐IA and control groups. The majority of patients in both groups had stage IV cancer. PD‐1/PD‐L1 inhibitors were the most frequently administered ICI regimen, used in 76.0% of the ICI‐IA group and 70.3% of the control group. More patients in the control group received concurrent chemotherapy, as compared with the ICI‐IA group (18.0% vs 9.0%). The same percentage of patients in both groups developed nonrheumatologic irAEs (32.0%).

The median observation time in CanRIO‐AB from ICI initiation was slightly longer in the ICI‐IA group compared to the control group (33.9 months vs 24.0 months, Table 2), although IQRs overlapped. Similarly, the provincial registry observation time from ICI initiation period was longer in the ICI‐IA cohort (34.4 vs 23.6 months), with overlapping IQRs. Despite the shorter observation period in the control group, the percentage of patients who died during the observation period was similar in both groups (39.0% vs 38.3%).

Of the 100 individuals with ICI‐IA, the median time between ICI initiation and IA onset was 3.8 months (IQR 1.7–9.7, Table 3). At presentation, the median number of tender joints was 4.5 (IQR 1–12), and median number of swollen joints was 4 (IQR 2–12). The most common pattern of joint involvement was oligoarthritis, observed in 45 patients (45.0%), with most patients having symmetrical joint involvement (60.0%). Seropositivity for either rheumatoid factor (9.0%) and anti‐cyclic citrullinated peptide antibodies (1.0%) was uncommon. The majority (64.0%) of ICI‐IA cases were classified as being CTCAE (Common Terminology Criteria for Adverse Events) grade 2 severity (ie, moderate symptoms that limit activities of daily living ²⁶ ), and most were treated with conventional synthetic DMARDs (csDMARDs; 78.0%) and/or glucocorticoids (68.0%). ICI therapy was held in 37% of patients, was discontinued in 35%, was continued in 4%, or had already been discontinued before ICI‐IA onset in 24%.

Table 3.

ICI‐IA phenotype and treatment^*

Variable	ICI‐IA
	n = 100
ICI initiation to symptom onset, median (IQR), months	3.8 (1.7–9.7)
Tender joints, median (IQR)	4.5 (1–12)
Swollen joints, median (IQR)	4 (2–12)
Pattern of joint involvement, n (%)
Oligoarthritis	45 (45.0)
Polyarthritis	41 (41.0)
Monoarthritis	14 (14.0)
Axial	1 (1.0)
Symmetric joint distribution, n (%)	60 (60.0)
Associated symptoms, n (%)
None	89 (89.0)
Bursitis	7 (7.0)
Tenosynovitis	2 (2.0)
Enthesitis	2 (2.0)
PMR onset symptoms	2 (2.0)
Inflammatory back pain	1 (1.0)
Achilles tendonitis	0 (0.0)
Patient global assessment (0–10 scale a ), median (IQR)	7.0 (5.0–8.0)
Physician global assessment (0–10 scale a ), median (IQR)	7.0 (5.7–8.0)
Morning stiffness, n (%)
≤1 h	37 (37.0)
>1 h	16 (16.0)
None	9 (9.0)
Positive RF, n (%)	9 (9.0)
Positive anti‐CCP, n (%)	1 (1.0)
CRP, median (IQR)	47.3 (11.1–112.2)
Joint irAE grading at its worst (CTCAE grade b ), n (%)
Grade 1	7 (7.0)
Grade 2	64 (64.0)
Grade 3	27 (27.0)
Grade ≥4	0 (0.0)
Joint symptoms improved by activity, n (%)	12 (12.0)
Inability to make a fist, n (%)	12 (12.0)
Decreased function, n (%)	20 (20.0)
Reduced range of motion hips, n (%)	5 (5.0)
Reduced range of motion shoulders, n (%)	10 (10.0)
Glucocorticoids, n (%)	68 (68.0)
Oral glucocorticoid	62 (62.0)
Intra‐articular glucocorticoids	5 (5.0)
IM glucocorticoids	1 (1.0)
IV glucocorticoids	0 (0.0)
Narcotics, n (%)	5 (5.0)
Conventional synthetic DMARDs, n (%)	78 (78.0)
Hydroxychloroquine	73 (73.0)
Methotrexate	27 (27.0)
Sulfasalazine	8 (8.0)
Leflunomide	3 (3.0)
Mycophenolate	0 (0.0)
Azathioprine	0 (0.0)
Other	0 (0.0)
Biologic DMARDs, n (%)	2 (2.0)
Targeted synthetic DMARDs, n (%)	0 (0.0)
ICI management after ICI‐IA, n (%)
ICI held	37 (37.0)
ICI discontinued	35 (35.0)
ICI continued	4 (4.0)
Not on ICI at ICI‐IA onset	24 (24.0)

^*Missing data: morning stiffness (38), RF (20), anti‐CCP (23), CRP (3), patient global at initial assessment (17), physician global at initial assessment (40), joint symptoms improved by activity (79), decreased function (38), inability to make a fist (36), reduced range of motion hips (28), and reduced range of motion shoulders (22). anti‐CCP, anti–cyclic citrullinated peptide antibodies; CRP, C‐reactive protein; CTCAE, Common Terminology Criteria for Adverse Events; DMARD, disease‐modifying antirheumatic drug; ICI, immune checkpoint inhibitor; ICI‐IA, immune checkpoint inhibitor–associated inflammatory arthritis; IM, intramuscular; IQR, interquartile range; irAE, immune‐related adverse event; IV, intravenous; PMR, polymyalgia rheumatica; RF, rheumatoid factor.

^a Patient and physician global assessments were recorded on a 0–10 scale: 0 = no disease activity, 10 = maximum disease activity. ⁴²

^b CTCAE: grade 1 = asymptomatic or mild symptoms; grade 2 = moderate symptoms that limit instrumental activity daily living; grade 3 = severe symptoms that limit self‐care activity daily living, persistent or significant disability; grade 4 = life threatening; grade 5 = death directly related to adverse event. ²⁶

Of the 228 patients in the study cohort, 14 had at least one IA ICD code registered in the three years before ICI initiation, all of whom were classified as negative for ICI‐IA by all of our ICD code–based case definitions. A total of 114 of 228 patients had at least one IA ICD code registered after ICI initiation, with a median time from ICI initiation to first IA code of 8.9 months (IQR 4.8–13.8). Of those who had more than one IA ICD code after ICI initiation, the median time between first and second IA ICD codes was 54 days (IQR 16–104). Among the 81 patients with two codes, 77 had the second code within one year of the first, 80 within two years, and one had an interval of more than three years between the first and second code. Among the 71 patients with three codes, 63 had the third code within one year of the first, 70 within two years, and one had an interval of more than three years between the first and third code. The specific IA codes registered at first, second, and third visits, by IA status, are shown in Supplementary Figure 1 and listed in Supplementary Table 2. The most frequently recorded IA code was 714.x.

The performance of the core ICI‐IA ICD code–based case definitions are presented in Table 4. Sensitivity estimates ranged from 1.0% (95% CI 0.0–3.0) to 88.0% (95% CI 81.6–94.4), specificity estimates ranged from 86.7% (95% CI 80.8–92.6) to 100.0% (95% CI 100.0–100.0), and C‐statistics ranged from 0.51 to 0.87 across all tested case definitions. The simplest case definition, requiring at least one PCD IA code had sensitivity of 88.0%, specificity of 86.7%, and C‐statistic of 0.87. Sensitivity estimates decreased and specificity estimates increased when additional PCD codes were required. The case definition requiring at least one DAD IA code demonstrated high specificity (100.0%, 95% CI 100.0%–100.0%) but very low sensitivity (1.0%, 95% CI 0.0%–3.0%). The addition of DAD codes to any of the PCD‐based case definitions did not enhance the performance of the corresponding PCD code–only case definition. Sex‐stratified performance estimates for the seven core case definitions were similar between men and women (Table 4).

Table 4.

Performance of ICI‐IA core case definitions with 95% confidence intervals for the full cohort and by sex^*

Case definition	Sensitivity	Specificity	PPV	NPV	C‐statistic
Full cohort, N = 228
≥1P	88.0 (81.6–94.4)	86.7 (80.8–92.6)	83.8 (76.8–90.9)	90.2 (85.0–95.5)	0.87 (0.83–0.92)
≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥3P	61.0 (51.4–70.6)	96.1 (92.7–99.5)	92.4 (86.0–98.8)	75.9 (69.3–82.5)	0.79 (0.73–0.84)
≥1H	1.0 (0.0–3.0)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	56.4 (49.9–62.8)	0.51 (0.49–0.51)
≥1H or ≥1P	88.0 (81.6–94.4)	86.7 (80.8–92.6)	83.8 (76.8–90.9)	90.2 (85.0–95.5)	0.87 (0.83–0.92)
≥1H or ≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥1H or ≥3P	61.0 (51.4–70.6)	96.1 (92.7–99.5)	92.4 (86.0–98.8)	75.9 (69.3–82.5)	0.79 (0.73–0.84)
Women, n = 99
≥1P	87.2 (76.7–97.7)	88.3 (80.2–96.5)	82.9 (71.4–94.4)	91.4 (84.2–98.6)	0.88 (0.81–0.94)
≥2P	74.4 (60.7–88.1)	95.0 (89.5–100.0)	90.6 (80.5–100.0)	85.1 (76.5–93.6)	0.85 (0.77–0.92)
≥3P	66.7 (51.9–81.5)	96.7 (92.1–100.0)	92.9 (83.3–100.0)	81.7 (72.7–90.7)	0.82 (0.74–0.90)
≥1H	2.6 (0.0–7.5)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	61.2 (51.6–70.9)	0.51 (0.49–0.54)
≥1H or ≥1P	87.2 (76.7–97.7)	88.3 (80.2–96.5)	82.9 (71.4–94.4)	91.4 (84.2–98.6)	0.88 (0.81–0.94)
≥1H or ≥2P	74.4 (60.7–88.1)	95.0 (89.5–100.0)	90.6 (80.5–100.0)	85.1 (76.5–93.6)	0.85 (0.77–0.92)
≥1H or ≥3P	66.7 (51.9–81.5)	96.7 (92.1–100.0)	92.9 (83.3–100.0)	81.7 (72.7–90.7)	0.82 (0.74–0.90)
Men, n = 129
≥1P	88.5 (80.5–96.5)	85.3 (76.9–93.7)	84.4 (75.5–93.3)	89.2 (81.7–96.8)	0.87 (0.81–0.93)
≥2P	63.9 (51.9–76.0)	92.6 (86.4–98.9)	88.6 (79.3–98.0)	74.1 (64.8–83.4)	0.78 (0.71–0.85)
≥3P	57.4 (45.0–69.8)	95.6 (90.7–100.0)	92.1 (83.5–100.0)	71.4 (62.1–80.7)	0.76 (0.70–0.83)
≥1H	–	–	–	–	–
≥1H or ≥1P	88.5 (80.5–96.5)	85.3 (76.9–93.7)	84.4 (75.5–93.3)	89.2 (81.7–96.8)	0.87 (0.81–0.93)
≥1H or ≥2P	63.9 (51.9–76.0)	92.6 (86.4–98.9)	88.6 (79.3–98.0)	74.1 (64.8–83.4)	0.78 (0.71–0.85)
≥1H or ≥3P	57.4 (45.0–69.8)	95.6 (90.7–100.0)	92.1 (83.5–100.0)	71.4 (62.1–80.7)	0.76 (0.70–0.83)

^*“–” denotes case definitions for which no patients met the case definition. H, Discharge Abstract Database code; ICI‐IA, immune checkpoint inhibitor–associated inflammatory arthritis; NPV, negative predictive value; P, Physician Claims Database code; PPV, positive predictive value.

Sensitivity analysis of additional case definitions criteria are shown in Table 5. The addition of the requirement that IA onset be within 365 days of ICI discontinuation to the core case definitions minimally reduced all performance measures. Restricting to rheumatoid arthritis ICD codes, or to only IA codes by either rheumatologists or internal medicine physicians, essentially did not change performance of the case definitions. However, restricting to only IA codes by rheumatologists increased specificity but reduced sensitivity across all case definitions. Inclusion of NACRS IA codes did not change the performance of any of the case definitions. Restricting multiple IA codes to being within one or two years of the first IA code reduced sensitivity without any improvement in specificity. The addition of PMR codes did not appreciably improve sensitivity but reduced specificity. The addition of glucocorticoid or DMARD prescription fill reduced sensitivity but slightly improved specificity.

Table 5.

Performance of additional criteria for ICI‐IA case definitions with 95% confidence intervals, N = 228^*

Case definition	Sensitivity	Specificity	PPV	NPV	C‐statistic
IA onset within 365 days after ICI discontinuation
≥1P	87.5 (80.9–94.1)	84.1 (77.9–90.3)	80.0 (72.3–87.7)	90.2 (85.0–95.5)	0.86 (0.81–0.90)
≥2P	67.7 (58.4–77.1)	91.7 (87.0–96.4)	85.5 (77.6–93.4)	79.6 (73.2–86.0)	0.80 (0.74–0.85)
≥3P	60.4 (50.6–70.2)	93.9 (89.9–98.0)	87.9 (80.0–95.8)	76.5 (70.0–83.1)	0.77 (0.72–0.83)
≥1H	–	–	–	–	–
≥1H or ≥1P	87.5 (80.9–94.1)	84.1 (77.9–90.3)	80.0 (72.3–87.7)	90.2 (85.0–95.5)	0.86 (0.81–0.90)
≥1H or ≥2P	67.7 (58.4–77.1)	91.7 (87.0–96.4)	85.5 (77.6–93.4)	79.6 (73.2–86.0)	0.80 (0.74–0.85)
≥1H or ≥3P	60.4 (50.6–70.2)	93.9 (89.9–98.0)	87.9 (80.0–95.8)	76.5 (70.0–83.1)	0.77 (0.72–0.83)
Rheumatoid arthritis ICD codes only (714.x, M05.x, M06.x)
≥1P	88.0 (81.6–94.4)	86.7 (80.8–92.6)	83.8 (76.8–90.9)	90.2 (85.0–95.5)	0.87 (0.83–0.92)
≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥3P	59.0 (49.4–68.6)	95.3 (91.7–99.0)	90.8 (83.7–97.8)	74.8 (68.2–81.5)	0.77 (0.72–0.82)
≥1H	1.0 (0.0–3.0)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	56.4 (49.9–62.8)	0.51 (0.49–0.51)
≥1H or ≥1P	88.0 (81.6–94.4)	86.7 (80.8–92.6)	83.8 (76.8–90.9)	90.2 (85.0–95.5)	0.87 (0.83–0.92)
≥1H or ≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥1H or ≥3P	59.0 (49.4–68.6)	95.3 (91.7–99.0)	90.8 (83.7–97.8)	74.8 (68.2–81.5)	0.77 (0.72–0.82)
IA code by rheumatologist only
≥1P	45.0 (35.2–54.8)	94.5 (90.6–98.5)	86.5 (77.3–95.8)	68.8 (61.9–75.6)	0.70 (0.64–0.75)
≥2P	23.0 (14.8–31.2)	97.7 (95.0–100.0)	88.5 (76.2–100.0)	61.9 (55.2–68.6)	0.60 (0.56–0.65)
≥3P	14.0 (7.2–20.8)	98.4 (96.3–100.0)	87.5 (71.3–100.0)	59.4 (52.8–66.0)	0.56 (0.53–0.60)
≥1H	1.0 (0.0–3.0)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	56.4 (49.9–62.8)	56.4 (49.9–62.8)
≥1H or ≥1P	45.0 (35.2–54.8)	94.5 (90.6–98.5)	86.5 (77.3–95.8)	68.8 (61.9–75.6)	0.70 (0.64–0.75)
≥1H or ≥2P	24.0 (15.6–32.4)	97.7 (95.0–100.0)	88.9 (77.0–100.0)	62.2 (55.5–68.9)	0.61 (0.56–0.65)
≥1H or ≥3P	15.0 (8.0–22.0)	98.4 (96.3–100.0)	88.2 (72.9–100.0)	59.7 (53.1–66.3)	0.57 (0.53–0.60)
IA code by rheumatologist or internal medicine physician only
≥1P	88.0 (81.6–94.4)	88.3 (82.7–93.9)	85.4 (78.6–92.2)	90.4 (85.2–95.6)	0.88 (0.84–0.92)
≥2P	67.0 (57.8–76.2)	93.8 (89.6–97.9)	89.3 (82.3–96.3)	78.4 (71.9–84.9)	0.80 (0.75–0.85)
≥3P	59.0 (49.4–68.6)	96.1 (92.7–99.5)	92.2 (85.6–98.8)	75.0 (68.4–81.6)	0.78 (0.72–0.83)
≥1H	1.0 (0.0–3.0)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	56.4 (49.9–62.8)	0.51 (0.49–0.51)
≥1H or ≥1P	88.0 (81.6–94.4)	88.3 (82.7–93.9)	85.4 (78.6–92.2)	90.4 (85.2–95.6)	0.88 (0.84–0.92)
≥1H or ≥2P	67.0 (57.8–76.2)	93.8 (89.6–97.9)	89.3 (82.3–96.3)	78.4 (71.9–84.9)	0.80 (0.75–0.85)
≥1H or ≥3P	59.0 (49.4–68.6)	96.1 (92.7–99.5)	92.2 (85.6–98.8)	75.0 (68.4–81.6)	0.78 (0.72–0.83)
Including NACRS IA codes as either P or H codes
≥1P	88.0 (81.6–94.4)	86.7 (80.8–92.6)	83.8 (76.8–90.9)	90.2 (85.0–95.5)	0.87 (0.83–0.92)
≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥3P	61.0 (51.4–70.6)	96.1 (92.7–99.5)	92.4 (86.0–98.8)	75.9 (69.3–82.5)	0.79 (0.73–0.84)
≥1H	2.0 (0.0–4.7)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	56.6 (50.2–63.1)	0.51 (0.50–0.52)
≥1H or ≥1P	88.0 (81.6–94.4)	86.7 (80.8–92.6)	83.8 (76.8–90.9)	90.2 (85.0–95.5)	0.87 (0.83–0.92)
≥1H or ≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥1H or ≥3P	61.0 (51.4–70.6)	96.1 (92.7–99.5)	92.4 (86.0–98.8)	75.9 (69.3–82.5)	0.79 (0.73–0.84)
Multiple IA codes must be within 1 y of first IA code
≥2P	65.0 (55.7–74.3)	93.8 (89.6–97.9)	89.0 (81.9–96.2)	77.4 (70.8–84.0)	0.79 (0.74–0.85)
≥3P	48.0 (38.2–57.8)	96.1 (92.7–99.5)	90.6 (82.7–98.4)	70.3 (63.5–77.1)	0.72 (0.67–0.77)
≥1H or ≥2P	65.0 (55.7–74.3)	93.8 (89.6–97.9)	89.0 (81.9–96.2)	77.4 (70.8–84.0)	0.79 (0.74–0.85)
≥1H or ≥3P	48.0 (38.2–57.8)	96.1 (92.7–99.5)	90.6 (82.7–98.4)	70.3 (63.5–77.1)	0.72 (0.67–0.77)
Multiple IA codes must be within 2 y of the first IA code
≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥3P	58.0 (48.3–67.7)	96.1 (92.7–99.5)	92.1 (85.4–98.7)	74.5 (67.9–81.2)	0.77 (0.72–0.82)
≥1H or ≥2P	68.0 (58.9–77.1)	93.8 (89.6–97.9)	89.5 (82.6–96.4)	78.9 (72.5–85.4)	0.81 (0.76–0.86)
≥1H or ≥3P	58.0 (48.3–67.7)	96.1 (92.7–99.5)	92.1 (85.4–98.7)	74.5 (67.9–81.2)	0.77 (0.72–0.82)
Addition of PMR codes
≥1P	88.0 (81.6–94.4)	75.8 (68.4–83.2)	73.9 (66.1–81.8)	89.0 (83.1–94.9)	0.82 (0.77–0.87)
≥2P	68.0 (58.9–77.1)	85.2 (79.0–91.3)	78.2 (69.5–86.8)	77.3 (70.4–84.2)	0.77 (0.71–0.82)
≥3P	61.0 (51.4–70.6)	88.3 (82.7–93.9)	80.3 (71.3–89.2)	74.3 (67.4–81.3)	0.75 (0.69–0.80)
≥1H	1.0 (0.0–3.0)	98.4 (96.3–100.0)	33.3 (0.0–86.7)	56.0 (49.5–62.5)	0.50 (0.49–0.52)
≥1H or ≥1P	88.0 (81.6–94.4)	75.8 (68.4–83.2)	73.9 (66.1–81.8)	89.0 (83.1–94.9)	0.82 (0.77–0.87)
≥1H or ≥2P	68.0 (58.9–77.1)	85.2 (79.0–91.3)	78.2 (69.5–86.8)	77.3 (70.4–84.2)	0.77 (0.71–0.82)
≥1H or ≥3P	61.0 (51.4–70.6)	88.3 (82.7–93.9)	80.3 (71.3–89.2)	74.3 (67.4–81.3)	0.75 (0.69–0.80)
Addition of at least one filled prescription of glucocorticoid or DMARD within 40 days (before or after) of the first code of ICI‐IA
≥1P	77.0 (68.8–85.2)	89.8 (84.6–95.1)	85.6 (78.3–92.8)	83.3 (77.1–89.6)	0.83 (0.79–0.88)
≥2P	63.0 (53.5–72.5)	93.8 (89.6–97.9)	88.7 (81.4–96.1)	76.4 (69.8–83.1)	0.78 (0.73–0.84)
≥3P	57.0 (47.3–66.7)	96.1 (92.7–99.5)	91.9 (85.2–98.7)	74.1 (67.4–80.8)	0.77 (0.71–0.82)
≥1H	1.0 (0.0–3.0)	100.0 (100.0–100.0)	100.0 (100.0–100.0)	56.4 (49.9–62.8)	0.51 (0.49–0.51)
≥1H or ≥1P	77.0 (68.8–85.2)	89.8 (84.6–95.1)	85.6 (78.3–92.8)	83.3 (77.1–89.6)	0.83 (0.79–0.88)
≥1H or ≥2P	63.0 (53.5–72.5)	93.8 (89.6–97.9)	88.7 (81.4–96.1)	76.4 (69.8–83.1)	0.78 (0.73–0.84)
≥1H or ≥3P	57.0 (47.3–66.7)	96.1 (92.7–99.5)	91.9 (85.2–98.7)	74.1 (67.4–80.8)	0.77 (0.71–0.82)

^*“–” denotes case definitions for which no patients met the case definition. DMARD, disease‐modifying antirheumatic drug; H, Discharge Abstract Database code; IA, inflammatory arthritis; ICD, International Classification of Diseases; ICI, immune checkpoint inhibitor; ICI‐IA, immune checkpoint inhibitor–associated inflammatory arthritis; NACRS, National Ambulatory Care Reporting System; NPV, negative predictive value; P, Physician Claims Database code; PMR, polymyalgia rheumatica; PPV, positive predictive value.

DISCUSSION

In this study, we evaluated the performance of ICD code–based administrative case definitions of ICI‐IA against rheumatologist‐confirmed ICI‐IA cases and noncases in the CanRIO‐AB database. We found that a simple case definition requiring at least one PCD IA ICD code has a good balance of sensitivity (88%) and specificity (87%) for identifying ICI‐IA, and performance of the various case definitions was similar between men and women. Higher specificity could be achieved with algorithms that required more than one PCD IA code (up to 96%), but with loss of sensitivity (lowest sensitivity 61%). The addition of hospital (DAD) and emergency department (NACRS) based codes did not improve the performance of PCD‐based case definitions. The algorithms requiring at least one code in DAD or NACRS data had extremely low sensitivity as expected, given that ICI‐IA is usually managed in the outpatient setting. Highest specificity was achieved when IA codes were restricted to rheumatologists.

The baseline characteristics of our ICI‐IA group were similar to other reported contemporaneous ICI‐user and ICI‐IA cohorts. The breakdown by tumor type in our cohort (majority being skin and lung cancer) was consistent with recent irAE studies reporting lung cancer and melanoma each making up roughly one‐third of the ICI users, and PD‐1/PD‐L1 being the most frequently used ICI. ²⁷ , ²⁸ The median time from ICI initiation to joint symptom onset in the ICI‐IA group of 3.7 months was similar to that reported by Cappelli et al (5 months), ²⁹ Roberts et al (6.5 months), ³⁰ and Chan et al (2.8 months). ³¹ Nearly all patients with ICI‐IA in our cohort were seronegative for rheumatoid factor and anti–cyclic citrullinated peptide, consistent with previous studies which have highlighted the seronegative nature of ICI‐IA. ⁸ , ²⁴ , ²⁹ , ³² , ³³ , ³⁴ The majority of patients with ICI‐IA in CanRIO‐AB received glucocorticoids and csDMARDs, most commonly hydroxychloroquine, consistent with the ICI‐IA treatment patterns reported in the full CanRIO database. ⁸ , ²⁴ Use of biologic DMARDs was very low in our cohort (2%), as compared with studies from other countries, likely reflecting, at least in part, drug‐coverage differences. ¹³ , ³³ , ³⁵

Although ICI‐IA case definitions have not been previously validated, ICD‐based case definitions for similar IAs such as rheumatoid arthritis and psoriatic arthritis have been evaluated. Two Canadian validation studies found that an algorithm requiring a single outpatient physician claim for rheumatoid arthritis yielded high sensitivity and specificity (both in the 90% range), similar to our findings. ³⁶ , ³⁷ They also found that hospital‐ and/or emergency department–based case definitions were not sensitive but highly specific for rheumatoid arthritis and that algorithms requiring multiple rheumatoid arthritis codes showed high specificity but sacrificed sensitivity. Another Canadian validation study examined the performance of ICD code–based case definitions for psoriatic arthritis and found that although all algorithms had excellent specificity (97%–100%), there was overall lower sensitivity (34%–72%) as compared with previous rheumatoid arthritis studies. ³⁸ In our study, the overall performance of ICD code–based case definitions of ICI‐IA seems to fall somewhere between those reported for rheumatoid arthritis and psoriatic arthritis and confirms that like other IAs, ICI‐IA is predominantly managed in the outpatient setting. Oftentimes, more specific case definitions, such as definitions requiring a minimum of two instances of a target IA ICD code, are used. However, ICI‐IA differs from rheumatoid arthritis or psoriatic arthritis in many aspects that may make a case definition requiring only a single instance of an IA ICD code favorable: it is not uncommon that ICI‐IA resolves; ICI‐IA is often managed by oncology, even after an initial visit with a rheumatologist, and patients with ICI‐IA may die of their cancer before a follow‐up visit with their rheumatologist.

There are currently no widely accepted or validated classification criteria for ICI‐IA, although this work is in progress. ³⁹ One major point of contention when defining ICI‐IA is how long after ICI cessation new onset IA can be attributed to ICI therapy and be classified as ICI‐IA. In our primary analysis, we included all rheumatologist‐confirmed cases of ICI‐IA in CanRIO‐AB, regardless of the timing of IA symptom onset in relation to ICI cessation. However, in anticipation of future classification criteria having exposure restriction, we also conducted a sensitivity analysis in which patients who developed ICI‐IA more than 365 days after ICI discontinuation were reclassified as not having ICI‐IA. This resulted in only a slight decrease in performance measures. The decline in sensitivity may be partially attributed to the observed approximate five‐month lag between patient‐reported IA symptom onset in CanRIO‐AB and subsequent first IA‐coded health care visit. This observed lag and corresponding attenuation in sensitivity should inform the use of ICD code–based definitions of ICI‐IA that are meant to reflect IA symptom onset in relation to ICI use. For example, if the clinical criteria for ICI‐IA requires that IA onset is within one year of ICI cessation, the ICD code–based case definition of ICI‐IA may require that the first IA code be coded within 18 months of ICI cessation to account for the lag observed between IA symptom onset and first IA‐coded visit.

In our core set of algorithms that required multiple IA codes, we chose to apply a liberal time restriction of maximum three years between codes because patients with ICI‐IA are often followed by oncologists, who are less likely to enter ICI‐IA codes versus cancer‐related codes for their visit diagnosis (even though in Alberta a physician can enter up to three diagnostic codes for any outpatient visit, most only enter one). Applying a more restrictive one‐year or two‐year time restriction between multiple IA codes only decreased sensitivity without increasing specificity. Although similar findings have been reported for rheumatoid arthritis, with sensitivity of 77% to 84% when a one‐year or two‐year time restriction was applied between codes, compared to a sensitivity of 90% if there was no time restriction, ³⁶ the drop in sensitivity was more pronounced for ICI‐IA. These findings highlight that strict time restrictions between multiple IA codes can result in missing true ICI‐IA cases, without improving specificity. Using rheumatoid arthritis–only ICD codes did not change the performance measures, so the core case definitions could be further simplified to only rheumatoid arthritis ICD codes without losing accuracy. The best PPVs were achieved when restricting to IA codes by rheumatologists, and thus, these additional criteria may be beneficial when identifying ICI‐IA cases from cohorts with low ICI‐IA prevalence.

Strengths of this study include the relatively large sample size of rheumatologist‐confirmed cases of ICI‐IA versus controls. Cancer diagnoses were ascertained from the Alberta Cancer Registry, which has been validated and received gold star certification from the North American Association of Central Cancer Registries, with more than 95% case ascertainment. ⁴⁰ , ⁴¹ ICIs are covered in Alberta under the provincial health plan and are administered in a hospital setting, ensuring a high ascertainment rate of ICI therapy at the population level. Similarly, physician, hospital, and emergency department visits are covered by the provincial health plan, and thus, all visits within the province are captured in the provincial health administrative databases. Using these databases, we were able to look back three years before the ICI initiation date to evaluate pre‐existing IA to ensure that our case definitions were able to discriminate those with a flare of pre‐existing IA from true ICI‐IA. We conducted a sex‐stratified analysis, although ICI‐IA does not appear to have a sex predilection unlike most rheumatic diseases, and the analysis showed that estimates of performance were similar between women and men, with overlapping 95% CIs, providing reassurance that our case definitions can be used across sexes.

We acknowledge that this study has limitations. The CanRIO‐AB database only includes patients referred to rheumatologists, which may bias our sample toward patients with better access to rheumatology care and/or those with persistent or more severe joint symptoms. Thus, the performance of the case definitions tested in this study may not generalize to ICI users who experience mild or transient joint symptoms or to those with limited access to rheumatologists. Additionally, the PPV and NPV estimates reported in this study will vary depending on the prevalence of ICI‐IA of the population to which these case definitions are applied. Similarly, the choice of the “best” case definition(s) will depend on whether sensitivity or specificity is the priority for a given use case. A recent study examined the incidence rate of ICI‐IA using various unvalidated ICD code–based case definitions and found that the incidence rate ranged from two to seven per 100 person‐years depending on the case definition used, ¹³ underscoring the importance of evaluating the attributes of all case definitions that are being used. Finally, although this study validated the case definitions using population‐level data in Alberta, external validation is required to assess the performance of these case definitions in other regions.

In conclusion, ICD code–based case definitions can accurately detect ICI‐IA in administrative health data, and a simple algorithm of at least one IA PCD code demonstrated a good balance of sensitivity and specificity in both sexes. However, the selection of the ICI‐IA ICD code–based case definition for use in a given study will depend on the use setting and the relative importance of optimizing PPV versus NPV. Development of a specific ICD code for ICI‐IA would likely improve the accuracy of ICI‐IA case detection using administrative health data, but in its absence, this study supports the use of currently available ICD code–based case definitions of ICI‐IA for disease surveillance and research using administrative health data.

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Ye confirms that all authors have provided the final approval of the version to be published and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

Supporting information

Disclosure form.

Table S1: Anatomical Therapeutic Chemical (ATC) codes used for sensitivity analysis including the additional criteria of at least one filled prescription of glucocorticoid or disease modifying anti‐rheumatic drugs (DMARD) within 40 days (before or after) of the first code of ICI‐IA

Table S2: Frequency of specific inflammatory arthritis (IA) codes recorded at each visit, by reference immune checkpoint inhibitor associated IA (ICI‐IA) status

Figure S1: Frequency of ICD codes ascertained, by visit and inflammatory arthritis (IA) status