Prediction tools for postoperative pain.

Achieving adequate pain control after surgery has been emphasized by many.[5] Poor pain control may lead to increased postoperative opioid utilization, development of persistent surgical pain, and extended hospital stays.[2,4] In the cross-sectional study by Schnabel et al.,[3] the authors used the PAIN-OUT registry containing 50,005 patients to elucidate 8 significant risk factors for severe acute postoperative pain (the numeric rating scale for pain ≥7) on postoperative day 1. Using a split-sample study design, they found patients with ≥3 risk factors were more likely to have severe postoperative pain, to spend more time in severe pain (>20%), and wished to have received more pain interventions.[3]

The usefulness of their prediction model lies in its ability to risk-stratify patients in the preoperative setting, so personalized treatment strategies can be developed to improve pain outcomes. As such, prognostic variables included in the prediction model should be attainable in the preoperative setting. However, the prediction model in this study included 2 risk factors (feeling anxious or helpless due to pain) that are postoperative characteristics, limiting their utility in predicting pain control in the preoperative setting. Furthermore, the authors included countries with median worse pain intensity of the numeric rating scale ≥7 as one of the prognostic factors, limiting the external validity of the model to participating countries in the PAIN-OUT registry. The authors addressed these limitations by creating a simplified risk score containing 4 risk factors (younger age, female sex, preoperative opioid use, and preoperative chronic pain). However, the ability of this score to separate those who will have severe pain from those who will have adequate pain control was poor (C-statistic of 0.61). This, unfortunately, limits the utility of the score in the clinical setting.

The reporting of their multivariable prediction model and risk score could be improved. The standard for reporting prognostic studies involving prediction models is the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement.[1] The performance of prediction models and scores in the training and validation data sets should be expressed in terms of calibration and discrimination. The authors failed to present either of these 2 model performance metrics for their gross training model and their 9-point risk score thereafter. Without this information, it is unclear whether there was significant erosion in the predictive performance after transformation to the risk score and prevented the internal validation of their findings using the validation data set.

Although it is tempting to develop prediction rules that are generalizable to all surgical specialties, they are unable to account for specialty specific factors that could impact postoperative pain. Including these risk factors (eg, number of spinal levels operated and fusion procedures in spine surgery) may improve a prediction rule's predictive performance.[4] In some cases, these risk factors are modifiable as surgeons have the ability to choose which surgical procedure to offer. A shared-decision approach may help educate patients about the pros and cons of various surgical options while taking into account their impact on postoperative pain.

Disclosures

The authors have no conflicts of interest to declare.