To the Editor:
Congratulations to Mazo et al.1 for their elaborate and extensive work. They designed the Prospective Evaluation of a RIsk Score for postoperative pulmonary COmPlications in Europe (PERISCOPE) study to improve the external validity of the Assess Respiratory Risk in Surgical Patients in Catalonia (ARISCAT) score, which they describe as the only prospective internal validated score to predict postoperative pulmonary complications.2 Therefore, they tested the generalizability of this score in a large European cohort and three subsamples: Spain, Western Europe, and Eastern Europe. They conclude that their risk score predicts three levels of postoperative pulmonary complications in an area outside the development setting.
Yet, to us the generalizability of the score seems doubtful, because the postoperative mortality reported in this study is inconsistent with the mortality reported in the European Surgical Outcomes Study (EUSOS).3
This inconsistency is conspicuous because the design, especially sampling strategies, of PERISCOPE resembles in many details the EUSOS study:
Both groups performed a multicenter design including numerous European hospitals.
Both groups defined continuous 7-day cohort periods to collect data of patients undergoing an in-hospital surgical procedure.
Both studies excluded patients undergoing obstetric procedures.
Both studies observed in-hospital mortality as an important outcome variable.
PERISCOPE (n = 5.099 patients) reports an overall in- hospital mortality of 0.9%, Spain (n = 2.000): 1.0%, Western Europe (n = 1.538): 0.8%, Eastern Europe (n = 1561): 0.9%. The crude mortality in the EUSOS study (n = 46.539 patients) was 4% ranging from 1.2% in the participating hospitals in Iceland (n = 162) to 21.5% in Latvia (n = 302). The United Kingdom provided the biggest sample of n = 10.630 patients, the mortality rate was 3.6%. In Spain (n = 5.433), 3.8% of surgical patients died.
This significant difference between both studies is especially remarkable due to the high-risk surgical procedures like cardiac or neurosurgery, which are included in the PERISCOPE but not in the EUSOS study. With respect to the aim of the PERISCOPE study, which is to improve generalizability, we consider it therefore indispensable to include this observation in the validation of the predictive score.
Two of the authors were involved in both publications. We wonder, why they did not discuss this important possible restriction of their validation study. Possibly the analysis of the ARISCAT study, in which the score to predict pulmonary complications was developed, gives an important clue to interpret the data.
Mazo et al.1 refer to the excellent internal validity of the ARISCAT study. Internal validity means optimal control of study conditions to ensure that the covariation of predictive score and outcome is not biased (nonspuriousness). Nonetheless, the better the internal validity is, the more limited is the external validity, i.e., the more elaborated the strategies are to control confounding influences, the more limited is the generalizability of a study.
The PERISCOPE study increases the external validity of the predictive score by a large degree of replication of the ARISCAT design in a new sample of patients. This strategy limits this generalizability to the special conditions as reported in the ARISCAT study. These conditions differ from the EUSOS investigation with high external and less controlled internal validity. Thus, the differences in mortality reported in the optimal controlled ARISCAT and PERISCAT studies compared with the EUSOS study may be explained simply by the effects induced into the participating hospitals by the studies itself. It is possible that the fact that respiration was studied directed the attention of hospital staff toward more careful observation of the respiration of postsurgical patients.
From our point of view, is it useful to consider this following aspect: if the authors conclude that increased attention to respiration (e.g., simply measuring oxygen saturation) may have contributed to reduce mortality, we will obtain a very easy to handle but highly effective approach to significantly reduce mortality in our hospitals.
The authors declare no competing interests.