Development of self-phenotyping tools to empower patients and improve diagnostics

ABSTRACT Deep phenotyping is important for improving diagnostics and rare diseases research and is especially effective when standardized using Human Phenotype Ontology (HPO). Patients are an under-utilized source of information, so to facilitate self-phenotyping we previously “translated” HPO into plain language (“layperson HPO”). Another self-phenotyping survey, GenomeConnect, asks patient-friendly questions that map to HPO. However, self-reported data has not been assessed. Since not all HPO terms are translated to layperson HPO or in the GenomeConnect survey, we created theoretical maximum-accuracy phenotype profiles for each disease for each instrument, representing the theoretical maximum performance. Both instruments performed well in analyses of semantic similarity (area under the curve 0.991 and 0.954, respectively). To explore the real-world implications, we randomized participants with diagnosed genetic diseases to complete the GenomeConnect, Phenotypr, or both instruments. For each diagnosed disease, we compared the derived disease profile to the patient-completed profile for each instrument. Profiles resulting from participant responses to the GenomeConnect survey were more accurate than to the Phenotypr instrument. The Phenotypr instrument had a tighter distribution of scores for respondents who did both instruments and was therefore more precise. We evaluated the ability of each known Mendelian disease HPO phenotype profile to retrieve the corresponding disease. We conducted interviews and generally participants preferred the GenomeConnect multiple choice format over the autocomplete Phenotypr format. Our results demonstrate that individuals can provide rich HPO phenotype data. These results suggest that self-phenotyping source of information could be used to support diagnostics or supplement profiles created by clinicians.