Abstract 1280P
Background
HAprog is a freely and user-friendly app for prognostication of advanced cancer outpatients. It is a computational transcription of two prognostic models: a ‘full version' composed of 6 parameters: sex, locoregional disease, sites of metastasis, ECOG-PS, WBC and albumin; and a 'clinical version' adding 'antineoplastic treatment' and without laboratory variables. The prognostic predictions are expressed in probability of being alive (in 30, 90 and 180 days) and expected scenarios (days).
Methods
A sample of advanced cancer outpatients were randomly selected from a retrospective databank cohort (2015-2019). HAprog was assessed at the time patients were first referred to palliative care. The clinical version was evaluated in all cases and the complete version when laboratory variables were available. Calibration (Hosmer-Lemeshow test) and discrimination (C-index and area under the receiver operating characteristic curve [AUROC]) were measured. Agreement analyzes were performed comparing predicted survival scenarios with the actual overall survival. Correlation analysis were performed using Spearman's test.
Results
HAprog clinical and full versions were evaluated in a total of 255 and 133 patients, respectively. The median overall survival (mOS) was 82 (95% CI 72-105) days. The C-index was similar in HAprog full and clinical versions (0.698 and 0.684, respectively). The AUROC were >0.75 at 30, 90 and 180 days in both versions and there were good calibration results. The absolute agreement rate between the actual survival time and the interval between expected scenarios (lower-upper) was 65.9% (95% CI 60.1-71.9%). The agreement rate between the actual survival time and the interval between the worst and best scenarios was 92.2% (95% CI 88.5-95.4%). The mean times of the HAprog expected scenarios were significant correlated with the actual patient survival (Rho=0.663, 95% CI 0.578-0.725).
Conclusions
The HAprog is an app for prognostication with adequate calibration and discriminations properties. It is able to provide prognostic information with good accuracy and correlation in order to assist the clinicians in decision making process.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
Barretos Cancer Hospital.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.