Background Patient reported outcome measures are comprised of either sets of questionnaires or patient global assessment based on visual analogue scale (VAS). These patient-reported outcome measures lack accuracy and/or clinical feasibility when comparing heterogeneous patient groups with different diseases, or when characterizing patients with systemic disease involving different organ systems.
Objectives Developing a clinical feasible patient-reported outcome measure based VAS assessment of different organ systems.
Methods Patients were asked to rate their health status in a 10cm VAS (0–100%) concerning their global health as well as of different organ systems, namely heart, lung, muscle and joints, gastro-intestinal, metabolic, uro-genital, skin, neuro-psychiatric, eyes and ears. All VA-scales were “anchored”. Patients were advised to rate their health status below 75% if they felt “medical action is needed”, they should rate the health status <50% in case of a “strong need for medical action” and <25% in case of a “medical emergency”.
336 patients from different outpatient clinics (cardiologic, pneumologic, gastro-intestinal, nephrologic, neurologic, dermatologic, rheumatologic, ophthalmologic and obesity outpatient clinic) as well as patients from internal emergency clinics and a general practitioner clinic were evaluated. Both, patients and the attending physicians completed the Popgen-OSSA. In addition the attending physician was asked to document ranking of the 5 most important diagnoses of the patient.
Statistical analysis was carried out using non-parametric testing. Furthermore, to predict main diagnoses based on patients's as well as physician's OSSA state-of-the-art machine learning tools, namely support vector machines (SVMs), were applied. To assess model performance multi-class AUC (area under the ROC curve) according to Hand and Till (2001) was estimated based on repeated cross validation (10 folds, 5 repeats), optimizing the SVM's hyperparamters using grid search.
Results The test showed a good reproducibility. With a mean percentage of 74±0.98 SE and 66±1.17 SE, respectively, the physicians OSSA rating was significantly higher than the rating of the patients (pwilcoxon<0.001). Models predicting main diagnoses were constructed and estimated to perform with multi-class AUCs of 63.5% and 73.4% based on patient's and physician's OSSA, respectively.
Conclusions In this preliminary trial with low sample size the Popgen-OSSA showed a good reproducibility and allowed a correct allocation of the patient's clinical problem to involved organ system by SVM analysis with multi-class AUC of up to 73.4%. These data merit further investigation and development of the Popgen-OSSA on larger patient cohorts.
David J. Hand and Robert J. Till (2001). A Simple Generalisation of the Area Under the ROC Curve for Multiple Class Classification Problems. Machine Learning 45(2), p. 171–186. DOI: 10.1023/A:1010920819831.
Disclosure of Interest R. Zeuner Grant/research support from: Pfizer, Novartis, UCB, U. Gsell: None declared, M. Hübenthal: None declared, S. Schreiber: None declared, A. Franke: None declared, M. Laudes Grant/research support from: Roche, Lilly, MSD, J. Schröder: None declared