Melanoma survival with classification and regression trees analysis: a complement for the communication of prognosis to patients.
2019
BACKGROUND: Prognostic factors in cutaneous melanoma are commonly evaluated by the Cox proportional hazard model. However, the interpretation of the effect of multiple variables is not straightforward. Classification and Regression Trees Analysis (CART), which allows a more friendly data evaluation, could be a valid integration of the message from Cox model. METHODS: The CART algorithm splits up data, creating a "tree" of groups of patients with different profiles for risk of death. Results are easy to interpret in clinical practice. A total of 2,692 patients with invasive cutaneous melanoma registered in Romagna (northern Italy) between 1993-2012 and followed-up until the end of 2013 were included. The Cox model and CART analysis were applied to sex, patient age, histological subtype, Breslow's tumour thickness, ulceration, site of disease, and Clark level. RESULTS: The CART analysis identified 15 categories which were collapsed into five classes with statistically different survival. The best prognostic group (10-year observed survival, 99.1%) included subjects with Breslow's thickness ≤ 0.78 mm and age 16-81 years. The worst prognostic group (10-year observed survival, 35.8%) comprised subjects with thickness ≥ 3.75 mm and age 16-96 years. According to the Cox model, patient age, histological subtype, Breslow thickness, ulceration, and site of disease had a significant independent prognostic value. CONCLUSIONS: CART and Cox models provided consistent results. CART seemed friendlier in its interpretation and it could facilitate the communication of risk.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
26
References
0
Citations
NaN
KQI