Background/Purpose:

Pulmonary hypertension (PH) contributes substantially to systemic sclerosis (SSc)-related morbidity and mortality. It tends to develop later in the disease, creating an opportunity for early risk stratification.

Previously published prediction models for PH have been based on cross-sectional data.

We set out to develop a model predicting PH development within 12 months, accounting for disease duration and utilising time-updated clinical characteristics, including serial measurements of lung function. We compare the use of most recent lung function results and serial changes in lung function over the preceding 4 years.

Methods:

We used data from a large unselected longitudinal cohort of SSc patients.

Sequential survival analyses with origins set at 6 consecutive landmark (LM) time-points, 12 months apart, starting at 60 months from disease onset were performed. The predictor variables included time-invariant characteristics (sex, subset and autoantibodies) and LM-specific information (age, presence of organ disease, FVC and DLCO, % predicted). Time to PH from the LMs was calculated with censoring at 12 months. Analyses were combined using a stratified Cox proportional hazards model, with each LM representing a stratum.

Results:

The study cohort consisted of 652 SSc patients. Of those 41.3% had diffuse SSc, 14.9% were male and the average age at disease onset was 48 years. Most patients (96%) either died during follow-up or were followed for over 10 years from disease onset. At the end of follow-up 13.3% of the subjects had developed PH.

The two final multivariable models both included the values at LM of age, presence of pulmonary fibrosis (PF) and antibody specificities (anti-U3RNP and anti-RNA polymerase (ARA)). Model 1 used most recent DLCO values, while Model 2 incorporated patient-specific intercept and slope of DLCO change over the 4 years prior to the LM (Table 1).

Both models were very similar, demonstrating that older age, ARA and anti-U3RNP positivity increase the hazard for PH development. Lower DLCO, both included as most recent measurement and as intercept and slope for the serial change in DLCO over the 4 years prior to LM, predicted increased hazard for PH. This effect was attenuated by presence of clinically-significant PF. None of the estimated effects varied between LM strata.

The two models had very similar fit and discrimination performance with C-index=0.88 for the model with most recent DLCO and C-index=0.87 for the one using intercept and slope for the linear change in DLCO.

Conclusion: