How to identify subgroups in longitudinal clinical data: treatment response patterns in patients with a shortened dental arch

Background When dental patients seek care, treatments are not always successful,that is pa-tients' oral health problems are not always eliminated or substantially reduced. Identifying these patients (treatment non-responders) is essential for clinical decision-making. Group-based trajectory modeling (GBTM) is rarely used in den-tistry, but a promising statistical technique to identify non-responders in particu-lar and clinical distinct patient groups in general in longitudinal data sets. Aim Using group-based trajectory modeling, this study aimed to demonstrate how to identify oral health-related quality of life (OHRQoL) treatment response patterns by the example of patients with a shortened dental arch (SDA). Methods This paper is a secondary data analysis of a randomized controlled clinical trial. In this trial SDA patients received partial removable dental prostheses replacing missing teeth up to the first molars (N = 79) either or the dental arch ended with the second premolar that was present or replaced by a cantilever fixed dental prosthesis (N = 71). Up to ten follow-up examinations (1-2, 6, 12, 24, 36, 48, 60, 96, 120, and 180 months post-treatment) continued for 15 years. The outcomeOHRQoL was assessed with the 49-item Oral Health Impact Profile (OHIP). Ex-ploratory GBTM was performed to identify treatment response patterns. Results Two response patterns could be identified - "responders" and "non-responders." Responders' OHRQoL improved substantially and stayed primarily stable over the 15 years. Non-responders' OHRQoL did not improve considerably over time or worsened. While the SDA treatments were not related to the 2 response patterns, higher levels of functional, pain-related, psychological im-pairment in particular, and severely impaired OHRQoL in general predicted a non-responding OHRQoL pattern after treatment. Supplementary, a 3 pattern approach has been evaluated. Conclusions Clustering patients according to certain longitudinal characteristics after treatment is generally important, but specifically identifying treatment in non-responders is central. With the increasing availability of OHRQoL data in clinical research and regular patient care, GBTM has become a powerful tool to investi-gate which dental treatment works for which patients.