Confirming Prior and Identifying Novel Correlates of Acute Calcium Pyrophosphate Crystal Arthritis

Abstract

Objectives:

To investigate previously identified and novel correlates of acute calcium pyrophosphate (CPP) crystal arthritis among well-characterized cases.

Methods:

In this case-control study, we identified cases of acute CPP crystal arthritis using a validated algorithm (positive predictive value 81%) applied in the Partners HealthCare electronic health record (EHR). Cases were matched to general patient controls on year of 1^st EHR encounter and index date. Pre-specified potential correlates included sex, race, and comorbidities and medications previously associated with CPPD/acute CPP crystal arthritis in the literature. We estimated odds ratios (OR) and 95% confidence intervals using conditional logistic regression models adjusted for demographics, comorbidities, medications prescribed in the past 90 days, healthcare utilization, and multimorbidity score.

Results:

We identified 1697 cases matched to 6503 controls. Mean age was 73.7 (SD 11.8) years, 56.7% were female, 80.8% White, 10.3% Black. All pre-specified covariates were more common in cases than controls. Osteoarthritis (OR 3.08), male sex (OR 1.35), rheumatoid arthritis (OR 2.09), gout (OR 2.83), proton pump inhibitors (OR 1.94), loop diuretics (OR 1.60), and thiazides (OR 1.46) were significantly associated with acute CPP crystal arthritis after full adjustment. Black race was associated with lower odds for acute CPP crystal arthritis compared to White race (OR 0.47).

Conclusion:

Using a validated algorithm to identify nearly 1700 patients with acute CPP crystal arthritis, we confirmed important correlates of this acute manifestation of CPPD. This is the first study to report higher odds for acute CPP crystal arthritis among males.

INTRODUCTION

Acute calcium pyrophosphate crystal arthritis, previously called pseudogout, represents a common cause of acute monoarthritis in adults. It characterizes the acute inflammatory manifestation of calcium pyrophosphate deposition (CPPD) disease, which can also present as asymptomatic chondrocalcinosis, osteoarthritis with CPPD, and chronic calcium pyrophosphate crystal inflammatory arthritis [1]. Prior studies of associations between CPPD and other conditions employed a validated algorithm for CPPD but did not focus on the acute inflammatory manifestation of this crystalline arthritis [2, 3]. The few large-scale studies of associations with acute CPP crystal arthritis employed a non-validated definition requiring just one diagnosis code from a general practitioner [4, 5].

Epidemiologic studies of correlates of CPPD or acute CPP crystal arthritis report consistent positive associations for several comorbidities with a pathophysiologic connection to calcium pyrophosphate metabolism or deposition, such as hyperparathyroidism, hypomagnesemia, and osteoarthritis [2, 4–6]. The literature conflicts regarding other metabolic disorders and medications, such as diabetes, rheumatoid arthritis, and bisphosphonates, and risk of CPPD or acute CPP crystal arthritis [2, 4, 5]. Prior studies employed different definitions of CPPD or acute CPP crystal arthritis and different analytic designs that may explain the conflicting results and provide rationale for further study.

Well-characterized, large cohorts of acute CPP crystal arthritis patients are necessary for confirming associated conditions and investigating long-term outcomes of this inflammatory crystalline arthritis. We previously developed an algorithm to accurately identify acute CPP crystal arthritis using electronic health record (EHR) data [7]. The algorithm leverages information captured in narrative notes, radiology reports, laboratory data including synovial fluid crystal analysis, and other structured and unstructured EHR data to achieve a positive predictive value (PPV) of 81% -- notably higher than the PPV of 22% for acute CPP crystal arthritis using billing codes alone [7]. The present study aimed to confirm previously identified correlates of CPPD and acute CPP crystal arthritis in the literature, and to test for novel associations with acute CPP crystal arthritis among well-characterized cases.

METHODS

Study population and data source

We performed a case-control study among patients in the Partners HealthCare Research Patient Data Repository (RPDR), 1991–2017. The Partners RPDR includes EHR data for >5 million patients from Brigham and Women’s Hospital, Massachusetts General Hospital, and affiliated practices [8]. Acute CPP crystal arthritis cases were identified by applying a validated algorithm with PPV 81% for this acute inflammatory manifestation of CPPD in the Partners RPDR [7]. Case index date was the earlier of 1^st positive synovial fluid calcium pyrophosphate crystal analysis or 1^st natural language processing concept of “pseudogout” in narrative notes. Cases were matched to up to 4 general patient controls not classified as acute CPP crystal arthritis by the algorithm from 1991–2017. Matching occurred on year of 1^st EHR encounter and index date. The control index date was the date of an encounter within 30 days of the matched case index date. All cases and controls were required to have age $\ge$50 at index date -- as acute CPP crystal arthritis is rare before age 50 -- and $\ge$2 encounters in the 365 days before index date. The Partners HealthCare Institutional Review Board approved all aspects of this study. The data underlying this article cannot be shared publicly due to the privacy of individuals included in the study. The data will be shared on reasonable request to the corresponding author.

Covariate assessment

Pre-specified covariates of interest included sex, race, and comorbidities and medications previously associated with CPPD or acute CPP crystal arthritis [2, 4, 5]. Race was extracted from the EHR according to self-reported US Census race categories. We required $\ge$2 ICD-9/10 codes from 1^st EHR encounter through index date to define each comorbidity, as using $\ge$2 codes achieves higher PPV than $\ge$1 ICD code and assessing comorbidities in all available data prior to index date is not expected to introduce bias [9, 10]. Supplement 1 provides billing codes employed for each comorbidity. We recorded diuretic, bisphosphonate, and proton pump inhibitor prescriptions for oral or intravenous use in the 90-day window before and including index date, with the rationale that 90 days represents a biologically plausible exposure period that might be associated with an episode of acute CPP crystal arthritis. Multimorbidity score was calculated using a validated method based on billing codes for 40 conditions [11].

Statistical analysis

We developed conditional logistic regression models to estimate univariable odds ratios (OR) and 95% CI for each covariate. A series of multivariable models was adjusted for age, sex, total number of encounters from 1^st EHR encounter through index date as a measure of healthcare utilization, multimorbidity score, and covariates with univariable OR p<0.1. To assess the independent contributions of comorbidities and medications that treat those comorbidities, we developed separate models adjusted for comorbidities only, medications only, and comorbidities plus medications. Stratified models explored differences in associations by sex and osteoarthritis status. A sensitivity analysis defined comorbidities using $\ge$1 ICD code, allowing assessment of conditions that were rare in our dataset and have been previously been associated with CPPD or acute CPP crystal arthritis. A second sensitivity analysis additionally adjusted for prescription NSAIDs and glucocorticoids in addition to the comorbidities and medications included in the primary model, to evaluate for attenuation of the effect estimate for proton pump inhibitors. We performed a Bonferroni correction accounting for the number of covariates in the fully-adjusted models such that p<0.003 was considered statistically significant. Analyses were performed using SAS v9.4 (SAS Institute, Cary, NC).

RESULTS

We identified 1697 acute CPP crystal arthritis cases with mean age 73.6 (SD 10.7) years and 47.5% male matched to 6503 controls with mean age 73.7 (SD 12.0) years and 42.2% male. Among the 1697 cases, 1091 (64.3%) had a synovial fluid crystal analysis positive for CPP crystals. The remaining 35.7% of cases were classified by the algorithm due to a combination of clinical and/or radiologic features. Hypertension (66.8% of cases, 46.5% of controls) and osteoarthritis (54.3% of cases, 21.2% of controls) were the most common comorbidities; gout had been diagnosed in 14.5% of cases and 3.1% of controls (Supplement 2). Proton pump inhibitors were the most frequent medications of interest (25.3% of cases, 9.3% of controls). Bisphosphonate prescriptions in the 90 days before and through index date were rare (3.9% of cases and 2.5% of controls). Healthcare utilization and multimorbidity score were greater in cases than controls.

Nearly all demographics, comorbidities, and medications demonstrated a significant univariable association with acute CPP crystal arthritis (Table 1). Odds of acute CPP crystal arthritis were significantly lower in Black patients (OR 0.47 [95% CI 0.37, 0.61]) and significantly greater in males (OR 1.35 [1.17, 1.55]); patients with osteoarthritis (OR 3.08 [2.67, 3.55]), rheumatoid arthritis (OR 2.09 [1.50, 2.90]), gout (OR 2.83 [2.19, 3.66]), or chronic kidney disease (OR 1.61 [1.22, 2.14]); and patients with recent prescription for proton pump inhibitor (OR 1.94 [1.61, 2.33]), loop diuretic (OR 1.60 [1.27, 2.01]), or thiazide (OR 1.46 [1.16, 1.85]) in a fully-adjusted model. Bisphosphonates were not significantly associated with acute CPP crystal arthritis. In models adjusted for comorbidities without medications, and adjusted for medications without comorbidities, effect estimates were similar to ORs in the fully-adjusted model.

Table 1.

Odds ratios (95% CI) for acute CPP crystal arthritis

Covariate	Univariate Models	Model 1 (comorbidities)	Model 2 (medications)	Model 3 (fully-adjusted)
Male	1.24 (1.12, 1.39)	1.33 (1.16, 1.52)	1.37 (1.21, 1.55)	1.35 (1.17, 1.55)
Race
White	1.00 (ref)	1.00 (ref)	1.00 (ref)	1.00 (ref)
Black	0.64 (0.52, 0.78)	0.47 (0.37, 0.61)	0.53 (0.42, 0.67)	0.47 (0.37, 0.61)
Other or unknown	1.10 (0.92, 1.33)	0.96 (0.77, 1.19)	0.90 (0.73, 1.11)	0.95 (0.76, 1.19)
Comorbidities
Osteoarthritis	4.71 (4.18, 5.32)	3.07 (2.66, 3.53)	--	3.08 (2.67, 3.55)
Hyperparathyroidism	3.32 (2.25, 4.91)	2.01 (1.19, 3.41)	--	2.04 (1.20, 3.49)
Hypomagnesemia	6.36 (2.81, 14.39)	2.56 (0.87, 7.59)	--	2.41 (0.78, 7.41)
Hypophosphatasia	2.76 (0.62, 12.37)	**	--	**
Hemochromatosis	6.00 (1.69, 21.26)	**	--	**
Rheumatoid arthritis	3.22 (2.46, 4.20)	2.16 (1.57, 2.99)	--	2.09 (1.50, 2.90)
Gout	5.57 (4.55, 6.83)	2.88 (2.24, 3.70)	--	2.83 (2.19, 3.66)
Diabetes, type 1 or 2	1.86 (1.63, 2.13)	1.02 (0.85, 1.21)	--	0.98 (0.82, 1.17)
Hypertension	2.42 (2.16, 2.72)	1.24 (1.07, 1.43)	--	1.16 (1.00, 1.35)
Coronary artery disease	2.57 (2.12, 3.13)	1.31 (1.01, 1.70)	--	1.24 (0.95, 1.61)
Chronic kidney disease	3.64 (2.97, 4.47)	1.76 (1.34, 2.32)	--	1.61 (1.22, 2.14)
Hypothyroidism	1.90 (1.62, 2.22)	0.98 (0.80, 1.20)	--	1.01 (0.82, 1.24)
Osteoporosis	2.12 (1.80, 2.50)	1.07 (0.86, 1.32)	--	1.06 (0.85, 1.33)
Medications
Proton pump inhibitor	3.46 (2.99, 4.00)	--	2.10 (1.77, 2.50)	1.94 (1.61, 2.33)
Loop diuretic	3.11 (2.61, 3.70)	--	1.63 (1.31, 2.01)	1.60 (1.27, 2.01)
Thiazide diuretic	1.80 (1.49, 2.16)	--	1.49 (1.20, 1.86)	1.46 (1.16, 1.85)
Bisphosphonate	1.55 (1.16, 2.08)	--	1.12 (0.80, 1.58)	1.05 (0.72, 1.51)

Conditional logistic regression models 1–3 are adjusted for age, sex, race, total number of healthcare encounters, and multimorbidity score

Additional adjustments as follows:

Model 1: adjusted for all comorbidities (defined by ≥ 2 ICD9/10 codes from 1^st EHR encounter through index date)

Model 2: adjusted for all medications

Model 3: adjusted for all comorbidities and all medications

^**Removed from model due to failure of convergence

Bold indicates p <0.003

In models stratified by sex, osteoarthritis was associated with greater odds of acute CPP crystal arthritis in males (OR 3.63 [2.73, 4.81]) than females (OR 2.46 [1.95, 3.09]) (Table 2). Gout was associated with two- to three-fold significantly increased odds of acute CPP crystal arthritis in models stratified by sex and models stratified by osteoarthritis, similar to the primary analysis. A sensitivity analysis defining comorbidities by $\ge$1 billing code identified effect estimates similar to the fully-adjusted primary analysis with the exception of gout (Supplement 3). When gout was defined by $\ge$1 code rather than $\ge$2 codes, gout prevalence nearly doubled among cases (26.5% vs. 14.5%) and the OR for acute CPP crystal arthritis was higher (gout OR 3.80 [3.10, 4.64]). In the sensitivity analysis additionally adjusting for prescription NSAIDs and glucocorticoids, the OR for PPI use (OR 1.46 [1.20, 1.77)] was attenuated compared to the OR in the primary analysis (OR 1.94 [1.61, 2.33]) but remained statistically significant after Bonferroni correction.

Table 2.

Stratified analyses: multivariable adjusted odds ratios (95% CI) for acute CPP crystal arthritis^*

	Female n=2852	Male n=2044	Osteoarthritis n=1361	No osteoarthritis n=3193
Male	--	--	1.46 (1.06, 2.02)	1.27 (1.05, 1.53)
Race
White	1.00 (ref)	1.00 (ref)	1.00 (ref)	1.00 (ref)
Black	0.57 (0.41, 0.79)	0.44 (0.24, 0.81)	0.36 (0.22, 0.61)	0.55 (0.39, 0.79)
Other or unknown	1.51 (1.06, 2.17)	0.53 (0.35, 0.82)	1.39 (0.83, 2.32)	0.71 (0.51, 0.98)
Comorbidities
Osteoarthritis	2.46 (1.95, 3.09)	3.63 (2.73, 4.81)	--	--
Hyperparathyroidism	4.31 (1.88, 9.86)	0.32 (0.06, 1.69)	6.33 (2.06, 19.44)	1.67 (0.69, 4.07)
Hypomagnesemia	11.85 (0.93, 151.84)	2.03 (0.34, 12.11)	2.78 (0.16, 47.24)	2.26 (0.54, 9.41)
Rheumatoid arthritis	2.19 (1.33, 3.63)	3.00 (1.38, 6.54)	1.96 (1.14, 3.36)	2.14 (1.14, 4.01)
Gout	2.63 (1.50, 4.61)	2.96 (1.89, 4.62)	2.81 (1.72, 4.60)	2.38 (1.55, 3.67)
Diabetes, type 1 or 2	0.89 (0.66, 1.20)	0.80 (0.57, 1.13)	0.82 (0.57, 1.17)	0.95 (0.73, 1.26)
Hypertension	0.96 (0.76, 1.23)	1.25 (0.94, 1.67)	1.36 (0.96, 1.92)	1.08 (0.88, 1.33)
Coronary artery disease	1.00 (0.57, 1.73)	1.01 (0.66, 1.55)	1.32 (0.77, 2.27)	0.99 (0.65, 1.51)
Chronic kidney disease	2.25 (1.32, 3.85)	1.64 (0.94, 2.84)	2.92 (1.47, 5.82)	2.14 (1.40, 3.28)
Hypothyroidism	0.95 (0.71, 1.26)	0.63 (0.36, 1.11)	0.95 (0.63,1.43)	0.94 (0.67, 1.33)
Osteoporosis	1.00 (0.75, 1.34)	0.70 (0.31, 1.61)	1.16 (0.77, 1.74)	0.99 (0.67, 1.48)
Medications
Proton pump inhibitor	2.14 (1.58, 2.89)	1.66 (1.15, 2.38)	2.33 (1.56, 3.47)	2.02 (1.53, 2.65)
Loop diuretic	1.23 (0.84, 1.82)	2.57 (1.63, 4.03)	1.45 (0.87, 2.43)	1.95 (1.39, 2.71)
Thiazide diuretic	1.28 (0.89, 1.83)	1.66 (1.02, 2.68)	1.25 (0.75, 2.11)	1.19 (0.84, 1.69)
Bisphosphonate	0.93 (0.56, 1.54)	1.35 (0.46, 3.98)	1.39 (0.62, 3.12)	0.97 (0.56, 1.67)

^*Conditional logistic regression models are adjusted for age, sex, race, total number of healthcare encounters, multimorbidity score, comorbidities defined by ≥2 ICD codes, and medications. Models stratified by sex are not adjusted for sex. Models stratified by osteoarthritis are not adjusted for osteoarthritis.

Bold indicates p <0.003

DISCUSSION

We identified two novel correlates and confirmed previous associations with acute CPP crystal arthritis, the acute inflammatory manifestation of CPPD, among nearly 1700 well-characterized cases matched to general patients. Male sex (35% increased odds) and Black race (53% lower odds) were associated with acute CPP crystal arthritis among patients identified via an EHR-based algorithm. Osteoarthritis, rheumatoid arthritis, gout, chronic kidney disease, proton pump inhibitors, loop diuretics, and thiazide diuretics were confirmed as correlates of acute CPP crystal arthritis.

Sex was not associated with CPPD in a meta-analysis of eight studies that included 5042 subjects (OR 0.89); however, six of the eight studies defined CPPD by radiographic chondrocalcinosis – not by the clinical manifestation of acute CPP crystal arthritis [1]. Similar prevalence of radiographic chondrocalcinosis in males and females does not necessarily translate to similar risk of developing acute CPP crystal arthritis. The 35% increased odds of acute CPP crystal arthritis among males in our study may be driven by osteoarthritis, as osteoarthritis was associated with significantly greater odds for acute CPP crystal arthritis in males (OR 3.63) than females (OR 2.46). It is also possible that males were more likely than females to be classified as acute CPP crystal arthritis by the algorithm, e.g. if males were more likely than females to undergo joint aspiration for identification of CPP crystals. This type of bias in the algorithm, if it existed, might have resulted in higher prevalence of male sex among cases than controls with a resulting higher odds ratio for acute CPP crystal arthritis among males.

Prior studies did not report whether race was evaluated as a potential correlate of acute CPP crystal arthritis, though the risk of gout – the most common crystalline arthritis – varies by race and ethnicity [12]. Black race was associated with half the odds of acute CPP crystal arthritis compared to White race in our study. While the pathophysiology of CPPD is not yet fully elucidated, we do not suspect a biologic explanation underpinning the lower odds of acute CPP crystal arthritis among Black patients in this study. Rather, it is possible that clinical suspicion for acute CPP crystal arthritis may have been systematically less common for Black patients, resulting in a lower chance of classification by our EHR-based algorithm that heavily relies on discussion of “pseudogout” in notes and/or demonstration of synovial fluid CPP crystals. Prior knowledge of associations between gout and Black race may have biased clinicians toward diagnosing gout rather than acute CPP crystal arthritis.

In the same vein, frequent consideration of gout on the differential diagnosis likely contributes to the positive association between gout and acute CPP crystal arthritis. Gout prevalence among cases exceeded 25% when gout was defined by $\ge$1 code compared to 14.5% when defined by $\ge$2 codes, which is closer to prior reports on co-existent synovial fluid CPP and monosodium urate crystals (approximately 5%) [13]. Diagnostic uncertainty on the path to acute CPP crystal arthritis diagnosis (e.g. $\ge$1 gout code) may partly explain the positive association between these types of crystalline arthritis in our study.

Our results largely confirm prior epidemiologic studies of CPPD and acute CPP crystal arthritis. A case-control study in the UK THIN database included 795 “pseudogout” cases, defined by 1 or more diagnosis code for “pseudogout” by a general practitioner, and identified significant associations with osteoarthritis (OR 2.91), hyperparathyroidism (OR 4.87), and loop diuretics (OR 1.35) [5]. Another case-control study in the UK Clinical Practice Research Datalink included 2011 “pseudogout” cases, defined by 1 or more diagnosis code for “pseudogout” by a general practitioner, and reported that oral bisphosphonate prescription within the preceding 60 days was associated with 43% greater risk for “pseudogout” [4]. We used a 90-day window for medications and included oral and intravenous prescriptions, but did not observe an association between bisphosphonates and acute CPP crystal arthritis. In a nationwide cross-sectional study of US military veterans, positive associations with CPPD – broadly defined and not specifically focused on its acute inflammatory manifestation — were noted for hyperparathyroidism (OR 3.35), gout (OR 2.82), osteoarthritis (OR 2.26), rheumatoid arthritis (OR 1.88), and hemochromatosis (OR 1.87) [2]. We required age $\ge$50 in attempt to limit misclassification of acute CPP crystal arthritis, but this requirement may also have excluded patients with hemochromatosis and other rare metabolic conditions previously associated with CPPD.

Proton pump inhibitor use was significantly associated with increased risk for acute CPP crystal arthritis in the primary analysis. Confounding by indication due to NSAID and/or glucocorticoid use is unlikely, as the OR for proton pump inhibitors remained significantly elevated in a sensitivity analysis additionally adjusted for these medications; however, we were unable to adjust for over-the-counter NSAID use. Proton pump inhibitors have been associated with increased risk for developing radiographic chondrocalcinosis, possibly by causing hypomagnesemia and decreased breakdown of pyrophosphate [14].

A major strength of this study is the inclusion of well-characterized cases identified by a validated algorithm leveraging EHR data to accurately identify acute CPP crystal arthritis. Limitations include its retrospective nature and the use of billing codes to identify comorbidities, though we improved PPV by requiring $\ge$2 codes for comorbidities. Osteoarthritis prevalence was likely underestimated by billing codes, as we did not include radiographs in the study definition of osteoarthritis. Hypophosphatasia and hemochromatosis, which have been associated with CPPD in the literature, were rare in our dataset (possibly because we excluded patients younger than 50) and multivariable models became unstable when they were included. We were able to estimate OR for these conditions in a sensitivity analysis with a more lenient definition ($\ge$1 code) for comorbidities. We adjusted for number of encounters prior to index date and comorbidity index, both of which were greater among cases than controls, but residual confounding remains possible. Misclassification of patients with other types of arthritis or other CPPD manifestations as acute CPP crystal arthritis was expected to occur in approximately 19% of patients per our algorithm PPV of 81%. As previously reported, we manually reviewed 100 randomly-selected cases; 85 had acute CPP crystal arthritis, nine had osteoarthritis with CPPD, four had possible acute CPP crystal arthritis, and two had crystal-proven gout [7]. We thus expect only very small percentage of gout patients were misclassified as acute CPP crystal arthritis.

CONCLUSION

We confirmed osteoarthritis, rheumatoid arthritis, gout, chronic kidney disease, proton pump inhibitors, loop diuretics, and thiazide diuretics as correlates of acute CPP crystal arthritis, the acute inflammatory manifestation of CPPD. We identified male sex as a potential novel correlate of acute CPP crystal arthritis.

Significance and Innovation.

• We confirmed correlates of acute CPP crystal arthritis among ~1700 cases identified via validated algorithm

• Male sex is a novel correlate of acute CPP crystal arthritis (adjusted OR 1.35)