We are grateful to Costantini and colleagues1 for the additional references,2,3 which were not included in our systematic review4 because our search finished in October 2016. The two additional references were indexed the following month. Both show somewhat higher sensitivity and substantially lower specificity than reported in our meta-analysis. The study by Gómez-Batiste and colleagues showed that the surprise question was positive in 79% (837/1059) of their entire cohort of high-risk patients, with a sensitivity of 93.7% but a very low specificity of 26.4% for 12-month mortality.2 Based on these numbers, if the surprise question were used to screen a large population and trigger any intervention with limited availability, resources could be stretched thin with little gain owing to the low positive predictive value (32%) and the small number of patients “screened out.”
We do not believe that the two errors mentioned by Costantini and colleagues affected our results. We recognized the error in the abstract from Moroni and colleagues5 and corrected it in our analysis (see Table 3 in our CMAJ article4). The publication by Feyi and colleagues6 had a follow-up of 18 months; therefore, if there was an additional death in their cohort after publication, it may not have affected the 12-month mortality. Furthermore, the data from Feyi and colleagues also represented only 1% of the total patients included in our meta-analysis; therefore, any changes would not meaningfully affect our overall findings.
We agree with the conclusions and rationale of Costantini and colleagues. Palliative care has an unmet need for better triggers based on unmet needs, and the prognosis for prognostication-based triggers appears grim.
Footnotes
Competing interests: None declared.