4CPS-192 Estimating the survival prognosis of patients with advanced gastrointestinal malignancy on home parenteral nutrition: a retrospective, monocentre study

Background and importance The initiation of home parenteral nutrition (HPN) in patients with advanced malignancy is a highly controversial topic. Guidelines recommend reserving this therapy for patients with an expected survival of longer than 2–3 months. Administering HPN in patients with a shorter survival probably has little benefit, while creating the risk of PN related complications. As HPN in advanced cancer patients is becoming increasingly common in our hospital, we wanted to investigate whether current practices are supported by the rational use of HPN. Aim and objectives Firstly, this study sought to investigate the proportion of patients with advanced cancer receiving HPN in our hospital, surviving for longer than 2–3 months. Furthermore, we wanted to investigate whether the application of survival prediction models could improve estimation of the length of patient survival. Material and methods Survival proportions of 250 patients with advanced gastrointestinal malignancy receiving HPN in our hospital during 2008–2016 were examined. Additionally, agreement was assessed between observed survival times and the current inhospital survival prediction method (ie, physician’s clinical judgement) or survival estimation by a published prediction nomogram. Moreover, through the use of multivariable logistic regression on variables gathered from the studied patient set, both a 2 and 3 month survival prediction model were constructed and validated. Results The results showed that a relatively low proportion of patients actually met the proposed survival criteria (65.2% and 46.4% for 2 and 3 month survival lengths, respectively). Concerning survival prediction, clinicians predominantly tended to overestimate survival length. Furthermore, application of the published nomogram did not improve survival prediction. Therefore, de novo 2 and 3 month survival prediction models were developed. The 2 month prediction model consisted of four variables: Karnofsky performance score (KPS), Glasgow prognostic score (GPS), gender and serum sodium, while the 3 month model consisted of three variables: KPS, GPS and serum urea. Validation of constructed survival prediction models in an independent set of 99 patients showed discriminatory abilities that were comparable, but not superior, to the results obtained with the aforementioned survival prediction nomogram. Conclusion and relevance This investigation showed that correct patient survival prediction remains an intrinsically difficult exercise. In order for our constructed models to have clinical utility, further improvement is needed, possibly through the inclusion of additional predictors for survival. References and/or acknowledgements No conflict of interest.