395MO - Prognostic models for survival predictions in advanced cancer patients: A systematic review and meta-analysis

Background

Prognostic models in predicting survival of patients with advanced cancer based on clinical and/or biochemical parameters have been shown to outperform predictions by clinicians. However, factors included and prognostic accuracy vary significantly. We conducted a systematic review to summarize and compare the model-based cancer prognostic tools for the palliative care setting.

Methods

A systematic search was conducted in Ovid Medline, Embase, CINAHL Ultimate, and Scopus databases to identify articles on validated prognostic models. Model characteristics, performance statistics, and quality of evidence were summarized and compared. Meta-analysis was done by pooling Harrel's C (C-index) to demonstrate the performance of different prognostic models, and by meta-regression to explore factors affecting model performance.

Results

Thirty-five prognostic models were analyzed. 11 models utilized only clinical parameters, 4 used purely biochemical parameters and 20 used both. 34 out of 35 studies were rated as high risk of bias primarily due to inadequate reporting of calibration statistics. While most models were internally validated (n=27), the Palliative Prognostic Index (PPI) and Palliative Prognostic Score (PaP) were externally validated across studies. For performance of predicting 30-day survival, the overall c-index is 0.74 [CI: 0.71 – 0.77, n=28]. For externally validated models, the pooled C-indices of PPI and PaP were 0.68 [CI: 0.62-0.73, n=6] and 0.76 [CI: 0.70-0.81, n=10], respectively. Some internally validated models, like the Objective Prognostic Index, showed a promising C-index of 0.89 [CI: 0.84 – 0.95, n=1]. Models utilizing objective biochemical factors as parameters, as opposed to those utilizing clinical factors only or a combination of both, showed superior performance (p = 0.01).

Conclusions

Amongst the models analysed, both PaP and PPI exhibit moderate prognostic accuracy in predicting short-term survival in patients with advanced cancer. Some emerging models show promising predictive properties but needs further external validation. Implementation research is warranted to examine the longer-term efficacy of model-based cancer prognostication in palliative care setting.

Drs. Wong Yuen Lung and Fung Mong Yung have equally contributed to the study.

Clinical trial identification

PROSPERO Registration ID: CRD4202340326.

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.