Background/Purpose: Crystal arthropaties, including gout and calcium pyrophosphate deposition disease (CPPD), are prevalent and burdensome conditions. The ALLSTAR model allows the identification of reliable causal, rather than correlational, links between a given set of examinations and the outcome, correcting for clinical confounders and multiple hypotheses testing. The strength of such links is measured through a different metric than p-value, that is the causal effect. : Our primary aim was to characterise the clinical presentation of crystal arthropaties in female patients and define causal predictors of a missing diagnosis through the deployment of a statistically reliable machine learning tool.

Methods: Adult patients presenting with acute mono-oligoarthritis attributed to gout, CPPD or unspecified crystal arthropathy were systematically enrolled in five rheumatologic centres. Clinical data were gathered through a cross-sectional online survey conducted between January 1, 2023, and November 15, 2023. Individuals with a pre-existing diagnosis of any other inflammatory arthropathy were excluded from the analysis. Consequently, a comparative analysis between female and male patients was conducted. We performed a traditional correlational analysis between the presence of performed examinations and the actual diagnosis of undefined crystal arthropathy. Subsequently, we ran ALLSTAR on the same data to assess the significance of previously found correlations against causal relations.

Results: A total of 399 patients were enrolled, including 136 (34%) females, and 263 males. Female patients were found to be significantly older (mean 67.1 years old vs. 63.8, p=0.008) and exhibited a higher prevalence of comorbidities (60.3% vs. 17.2%, p< 0.001). The rate of women who underwent synovial fluid examination (36.8% vs. 18.4%, p=0.001) and X-ray imaging (XR 43.4% vs. 28.7% p=0.004) was significantly higher. However, in 37.5% of cases, female patients did not receive a definitive diagnosis, vs 14% in males. Different models of logistic regression found a plethora of predictors significantly associated with the diagnosis of undefined crystal arthropathy, such as female sex, older age, comorbidities, XR, US scans, and various XR and US findings. Conversely, applying the ALLSTAR tool, we highlight that only the absence of deposits at US examination is the strongest and only causal predictor of an undefinite diagnosis (effect 0.0403). In men, the only predictor of the undefinite diagnosis is the presence of effusion in US scans, even if displaying a much weaker effect than in women (five times weaker, effect 0.0079).

Conclusion: In this multicentre cohort, female patients with crystal arthropathy are frequently complex elderly individuals with comorbidities and a low quality of life. Despite undergoing various examinations, nearly 40% of cases do not receive a definite diagnosis. Applying a causal inference tool we demonstrate that the best predictor for an undefinite diagnosis is the absence of deposits in US scans.

« Back to ACR Convergence 2024

ACR Meeting Abstracts - https://acrabstracts.org/abstract/differences-between-correlation-and-causal-inference-in-detecting-predictors-of-undefined-crystal-arthropathy-in-women/