Development and validation of a multivariable prediction model for the early prediction of chronic postsurgical pain. Response to Br J Anaesth 2022; 129: e155

Editor—We thank Wang and Chen1Wang Y.S. Chen D.X. Development and validation of a multivariable prediction model for early prediction of chronic postsurgical pain in adults: a prospective cohort study. Comment on.Br J Anaesth. 2022; 129: e155Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar for their comments on our study.2van Driel M.E.C. van Dijk J.F.M. Baart S.J. et al.Development and validation of a multivariable prediction model for early prediction of chronic postsurgical pain in adults: a prospective cohort study.Br J Anaesth. 2022; 129: 407-415Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar The key concern raised is that the predictive performance of the models might decrease over time when multimodal analgesia and surgical techniques change. We agree that changes in multimodal analgesia and surgical techniques will influence pain on the first postsurgical day. Our model based on data collected during this first postsurgical day should therefore prospectively be optimised in future studies. However, we do not expect that changes in multimodal analgesia and surgical techniques will change our best performing model based on data from the 14th day after surgery. At most, fewer patients will be identified as having an increased risk on chronic postsurgical pain. We do hope that the high incidences of patients at risk of chronic postsurgical pain identified with our model will eventually lead to improvements in perioperative analgesia and surgical techniques. We accounted for differences between clinical practices as the robustness of the models has been tested by external validation. Model performance was quantified by making predictions from the original models and comparing these predictions with the actual outcomes in patients from the Erasmus University Medical Center. Despite differences in the surgical case mix between the development and validation data sets, external validation showed similar performance. However, there is a potential limitation concerning generalisability. As patients received surgery at two university medical centres, the reproducibility of our findings in non-university hospitals is unknown and requires additional validation. A second concern was that the predictors age, BMI, and pain scores were coded as continuous variables rather than converting the predictor into a dichotomous form for analysis. Categorisation by using one or more cut points may simplify the analysis and makes it easier for clinicians to use the predictor. It is important to recognise, however, that categorisation of continuous predictors that go into the model is unnecessary for statistical analysis and comes at the expense of losing valuable information. The information loss is greatest when the predictor is dichotomised (i.e. the continuous predictor is converted into categorical form with two categories using solely one cut point). Moreover, it is well known that the model's predictive performance can vary if different cut points are used for splitting. In the absence of a priori clinical consensus for a cut point and the recommendation to avoid data-driven approaches for cut point selection, there is a problem with specifying accurate cut points. Therefore, dichotomisation of age, BMI, and pain scores is considered as statistically inefficient and strongly discouraged.3Royston P. Altman D.G. Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea.Stat Med. 2006; 25: 127-141Crossref PubMed Scopus (1536) Google Scholar, 4van Walraven C. Hart R.G. Leave ‘em alone—why continuous variables should be analyzed as such.Neuroepidemiology. 2008; 30: 138-139Crossref PubMed Scopus (74) Google Scholar, 5Vickers A.J. Lilja H. Cutpoints in clinical chemistry: time for fundamental reassessment.Clin Chem. 2009; 55: 15-17Crossref PubMed Scopus (24) Google Scholar, 6Bennette C. Vickers A. Against quantiles: categorization of continuous variables in epidemiologic research, and its discontents.BMC Med Res Methodol. 2012; 12: 21Crossref PubMed Scopus (268) Google Scholar, 7Dawson N.V. Weiss R. Dichotomizing continuous variables in statistical analysis: a practice to avoid.Med Decis Making. 2012; 32: 225-226Crossref PubMed Scopus (128) Google Scholar, 8Collins G.S. Reitsma J.B. Altman D.G. Moons K.G.M. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement.BMJ. 2015; 350: 7594Crossref PubMed Scopus (1639) Google Scholar, 9Moons K.G.M. Altman D.G. Reitsma J.B. et al.Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration.Ann Intern Med. 2015 Jan 6; 162: W1-W73https://doi.org/10.7326/M14-0698Crossref PubMed Scopus (1595) Google Scholar Continuous predictors should ideally be kept as continuous. When keeping variables continuous, a linear predictor–outcome relationship is assumed, in which the predictor–outcome relationship does not differ substantially from the unknown ‘true’ relationship.8Collins G.S. Reitsma J.B. Altman D.G. Moons K.G.M. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement.BMJ. 2015; 350: 7594Crossref PubMed Scopus (1639) Google Scholar, 9Moons K.G.M. Altman D.G. Reitsma J.B. et al.Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration.Ann Intern Med. 2015 Jan 6; 162: W1-W73https://doi.org/10.7326/M14-0698Crossref PubMed Scopus (1595) Google Scholar This assumption can be checked by using the restricted cubic spline function. However, no procedure for simultaneously selecting predictors and functional forms has yet found wide acceptance.8Collins G.S. Reitsma J.B. Altman D.G. Moons K.G.M. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement.BMJ. 2015; 350: 7594Crossref PubMed Scopus (1639) Google Scholar, 9Moons K.G.M. Altman D.G. Reitsma J.B. et al.Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration.Ann Intern Med. 2015 Jan 6; 162: W1-W73https://doi.org/10.7326/M14-0698Crossref PubMed Scopus (1595) Google Scholar Therefore, we deliberately chose not to use the restricted cubic spline function for model development. In summary, we believe that the two models we presented in our recent paper2van Driel M.E.C. van Dijk J.F.M. Baart S.J. et al.Development and validation of a multivariable prediction model for early prediction of chronic postsurgical pain in adults: a prospective cohort study.Br J Anaesth. 2022; 129: 407-415Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar are robust in the academic medical centre setting to alert clinicians to undertake further assessment of patients at risk for chronic postsurgical pain. The point that the model should be prospectively optimised and periodically updated is well taken. Prediction model development represents a continuous process of updating and validating. The authors declare that they have no conflict of interest. Development and validation of a multivariable prediction model for early prediction of chronic postsurgical pain in adults: a prospective cohort study. Comment on Br J Anaesth 2022; 129: 407–15British Journal of AnaesthesiaVol. 129Issue 6PreviewEditor—van Driel and colleagues1 developed and validated a prediction model for chronic postsurgical pain (CPSP) in adults. The rationale for the use of predictive modelling in clinical risk scenarios is compelling, and the authors developed a model with four items that are readily available in clinical practice, which is a welcome advance. Despite the clinical relevance of the topic and the rigorous study design, certain issues require clarification. Full-Text PDF Open Archive