Background: The factors necessitating the need for referrals for in-person evaluations by a dermatologist are not adequately understood and have not been studied using automated text mining so far. The objective of this study was to compare the prevalence of required in-person dermatologist care in the presence or absence of certain clinical features. Methods: Observational cross-sectional study of 11,661 teledermatology reports made from February 2017 to March 2020. Results: The need for dermoscopy was associated with a 348% increase in the possibility of referral for in-person dermatologist evaluations (prevalence ratio [PR]: 4.48, 95% confidence interval [CI]: 4.17-4.82). Infectious diseases were associated with a 64% lower possibility of referral (PR: 0.36, 95% CI: 0.30-0.43). Discussion: Some lesions and poorly documented cases are challenging to assess remotely. This study presents a different approach to research more detailed data from teledermatology reports, using text mining, and points out the risk magnitude for demanding dermatologic in-person care of which feature analyzed. As limitations, the variables related to lesion location, size, and extension were not analyzed and the dictionaries used were originally in Brazilian Portuguese. Conclusions: Teledermatology seems sufficient for the management of 75% of clinical cases, especially acute in young patients with inflammatory or infectious lesions. Referrals for in-person dermatologist consultations were not only strongly associated with the need for dermoscopy, but also for therapeutic reasons like surgical procedures, phototherapy, and the use of some systemic medications.
Keywords: data mining; decision tree; dermatology; referral criteria; teledermatology.