Article Text

Download PDFPDF

OP0363 Optimisingprecision medicine by using genetics to assign diagnostic prior probabilities to patients with synovitis – proof of principle
  1. R. Knevel1,2,3,
  2. C. Terao4,5,
  3. J. Cui2,
  4. K. Slowikowski3,6,
  5. T. Huizinga1,
  6. B. Karlson2,
  7. K. Liao2,
  8. S. le Cessie7,
  9. S. Raychaudhuri3,8,9
  1. 1Rheumatology, LUMC, Leiden, Netherlands
  2. 2Rheumatology, Brigham and Women’s Hospital, Harvard Medical School, Boston
  3. 3Broad Institute, Cambridge, USA
  4. 4Rheumatology, University of Shizuoka, Shizuoka
  5. 5RIKEN Center for Integrative Medical Sciences, Kanagawa, Japan
  6. 6Genetics, Brigham and Women’s Hospital, Harvard Medical School, Boston, USA
  7. 7Epidemiology, LUMC, Leiden, Netherlands
  8. 8Rheumatology/Genetics, Brigham and Women’s Hospital, Harvard Medical School, Boston
  9. 9Rheumatology, university of Manchester, Manchester, USA


Background In patients with synovitis, the question is ‘Which disease does this patient have?’ However, traditional tests often only inform us about disease presence yes/no and disease discriminating symptoms often take a while to arise. Time independent information, such as genetics, might accelerate the diagnostic process. As increasing number of patients have genotyping data available in medical records prior to their visit, the question emerges: can genetic data facilitate disease differentiation in early disease?

Objectives Proof of principle study to test the differentiating ability of genetic profiles in patients with synovitis.

Methods We studied the most common rheumatologic diseases: rheumatoid and psoriatic arthritis, SLE, spondyloarthropathy and Gout. The population level disease probability for each disease comprised a sex adjusted disease prevalence and a weighted genetic risk score comprised of risk SNPs’ odds ratio from literature. Within case genetic probabilities (GProb) were obtained through normalisation of the population risk assuring a patient’s total disease probability of 1. So, each patient got a probability for each disease. GProb was developed in a simulated dataset and tested in

  1. Validation dataset of 1,211 rheumatology cases identified with ICD codes from 62,512 patients

  2. Replication dataset of 248 rheumatology cases identified by chart–review from 15,047 patients

  3. Clinical setting of prospective selected patients that presented with synovitis at the rheumatology outpatient clinic (n=242). Here, GProb was calculated for the five diseases plus the category ‘Other’.

Having multiple GProbs for each patient, we tested whether the GProbs referring to the patient’s real disease were higher than those that referred to the other phenotypes.

We used multinomial logistic regression with the six diseases as the dependent variables to test the additive value of GProb on top of clinical information.


  1. There was a strong significant correlation between GProb and the disease status (r=0.27 P<0.0001) with an AUC of 0.68.

  2. We observed a higher correlation with disease status in the more precisely identified cases (r=0.49 P<0.0001) and a high AUC 0.82

  3. Also in a prospective setting, the GProb performed well (P<0.0001 AUC 0.74 figure 1) especially in ruling out diseases (table 1).

ResultsThe clinical information alone explained 41% of the variance in the final diagnosis. Adding GProb significantly improved the predictive value (expl. variance increased to 51% p=0.0008).

Sensitivity analysis showed that the results were not driven by one disease.

Abstract OP0363 – Table 1

Performance of G-Prob in ruling out diseas in the prospective dataset (n=242)

Abstract OP0363 – Figure 1 Ability of GProb to differentiate six disease outcomes (RA, PsA, SLE, SpA, Gout, Other) in a prospective selection of patients with synovitis.

The graph depicts the mean GProb (with range) of each quintile of GProbs on the x-axis and the corresponding proportion of those GProbs that matched with the patient’s real disease on the y-axis. The graph demonstrates that the genetic probability of the disease highly resembles the real disease risk. In case of a perfect test performance, the dots would lie exactly on the diagonal (dashed) line.

Conclusions This study developed methodology for disease-discriminating tests.

In patients with synovitis, genetic data can facilitate decision making in early disease by ruling out and pointing towards the most likely phenotype. Seeing the increasing importance of an early diagnosis in patients with synovitis, genetics can be considered as part of a patient’s medical history.

Additional prospective studies will further need to validate this proof of principle study.

Disclosure of Interest None declared

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.