New Score Uses Intra-op Variables To Predict Respiratory Complications

Author: Michael Vlessides

Anesthesiology News

Investigators have been able to significantly improve the predictive abilities of the Score of Prediction of Postoperative Respiratory Complications, or SPORC, by incorporating intraoperative variables, an evolution that ultimately may improve patient care.

“Postoperative respiratory complications are common, leading to increased costs and health care resource utilization, and may be associated with increased mortality and adverse discharge disposition,” said Charlotte Lukannek, CandMed, a medical student at Beth Israel Deaconess Medical Center (BIDMC) and Harvard Medical School, both in Boston. “To minimize the risk of respiratory failure after surgery, the Agency for Healthcare Research and Quality recommends the assessment of risk factors to better prepare physicians to anticipate potential postoperative adverse events, as well as improve the allocation of health care resources after surgery.”

Although researchers at the institution and Massachusetts General Hospital had previously developed the SPORC prediction tool for postoperative respiratory complications (Anesthesiology2013;118[6]:1276-1285), the tool only includes preoperative variables. “We’ve learned from several studies at Massachusetts General Hospital and BIDMC that potentially preventable risk factors also arise during surgery, and we wanted to include these in our prediction instrument,” Lukannek explained.

“We are in a unique situation where we can utilize data from two independent health care network[s],” Lukannek said. “So we created the score in one network and externally validated it in the second independent network.”

The analysis used hospital registry data from 157,939 adult inpatients at both institutions between 2005 and 2017. All patients underwent noncardiac surgery with endotracheal intubation, and were extubated in the OR. The primary outcome was reintubation within the first three postoperative days requiring unplanned mechanical ventilation in the OR, PACU or ICU.

Identifying New Predictors

As the investigators reported at the 2018 annual meeting of the American Society of Anesthesiologists (abstract A2268), 90,893 patients were included in the score development cohort. Among them, reintubation within three days occurred in 699 cases (0.8%).

With respect to model development, the researchers began with the five variables from the original SPORC model: ASA physical status greater than III, emergency surgery, high-risk surgery (vascular surgery, thoracic surgery, neurosurgery or burn surgery), history of chronic pulmonary disease, and history of congestive heart failure.

“Based on recent publications, we also considered a host of intraoperative candidate variables and procedure-related risk factors that we knew could potentially increase the risk of early reintubation,” Lukannek explained.

Multivariable stepwise backwards regression and bootstrap resampling were used to identify the final predictors of postoperative reintubation. Net reclassification improvement and C-statistics were used to compare the new prediction model with its predecessor, and the model was validated in the external validation cohort.

The final model included five preoperatively available predictors: ASA physical status of III or greater, history of heart failure, history of chronic pulmonary disease, emergency surgery, and high Procedure Severity Index. The model also included seven intraoperative predictors: early post-intubation desaturation (blood oxygen saturation <90%), extended duration of surgery of 140 minutes or more and 225 minutes or more, high median fraction of inspired oxygen of greater than 0.61, high vasopressor dose of more than 0.18 mg norepinephrine equivalent dose, blood transfusion, absence of volatile anesthetic use, and absence of protective ventilation (driving pressure >15 mm Hg).

The model yielded an area under the receiver operating characteristic curve (AUC) of 0.84 (95% CI, 0.82-0.85). A Hosmer-Lemeshow test was not significant (P=0.46), indicating good model calibration. Most importantly, external validation confirmed good discriminative ability (AUC, 0.76; 95% CI, 0.73-0.77).

“The promising observation we made here is that even though the external validation cohort was characterized by a different case mix, severity and risk factor composition, we still found a promisingly high predictive value and good model discrimination,” Lukannek said.

With respect to performance, risk categories derived from the new prediction instrument—dubbed “SPORC II” by the investigators—showed greater accuracy than the original SPORC in a net reclassification. Indeed, 232 patients in the development cohort were adequately reclassified into a higher risk category while 16,209 patients were adequately reclassified into a lower risk category. Receiver operating characteristic comparison analysis showed a higher AUC for SPORC II (0.85; 95% CI, 0.83-0.86) than for the first-generation model (0.76; 95% CI, 0.74-0.78).

“In summary, we developed a new prediction model for early reintubation after surgery that utilized anesthesia- and surgery-related risk factors,” Lukannek said. “By the intraoperative data, we were able to create a substantially improved prediction model.”

Clinical Applications Near at Hand

“Many other risk scores use prospective validation, but yours is retrospective,” said Carlos Guerra Londono, MD, an anesthesiology resident at Henry Ford Hospital in Detroit. “Do you think you need prospective validation?”

“Honestly, I don’t think there is another score that’s validated in a totally independent cohort,” replied study senior author Matthias Eikermann, MD, PhD, a professor of anesthesia at Harvard Medical School and BIDMC. “I think what’s most important is that you have a different case mix and different risk factor composition in the other validation cohort, which we clearly have.”

Roman Schumann, MD, a professor of anesthesiology at Tufts Medical Center, in Boston, asked how the authors see SPORC II being used clinically. “It’s interesting and obviously important. I could see this easily leading to more research in large cohorts of patients. But do you see a clinical application for the average practitioner?”

“If you look at the statistics, as many as 30% patients may need reintubation,” Dr. Eikermann replied. “So you really want to make sure that everything is OK with these patients. You just don’t want to close your eyes and say, ‘We’re fine.’ At a minimum, you can use the scale and then decide what to do with the information. But it might change behaviors.”

“Where I can immediately see it helping is with respect to decision support,” Dr. Schumann said. “You might even have something pop up on the intraoperative [electronic health record] anesthesia screen which prompts practitioners to reconsider their postoperative management.”