Improving Large Data Research: An Analysis Of Comorbidity Indices And Approaches To Missing Values

The last ten years has seen the development and subsequent adoption of large and national datasets in orthopaedic surgery research. While these patient cohorts certainly have a strong potential to offer advancements to the field as a whole, there are concerns regarding the identification and subsequent implementation of the most appropriate methods of utilizing these data sources to produce the highest quality research. In this context, the objective of this thesis is two-fold. (1) To compare the discriminative ability for adverse outcomes of three different methods of measuring the totality of a patient\u27s health: the American Society of Anesthesiologist physical status classification system (ASA), the modified Charlson Comorbidity Index (mCCI), and the modified Frailty Index (mFI); (2) To evaluate the impact of a newer and accepted method of handling missing values, multiple imputation, in comparison with what is currently a commonly reported approach in major orthopaedic journals, complete case analysis. The following methods were used. (1) For Objective 1, patients undergoing posterior lumbar fusion surgeries were identified in the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) and the aforementioned patient comorbidity indices were calculated and assessed for their discriminative ability in predicting commonly studied perioperative adverse outcomes using an area under the curve analysis from the receiver operating characteristic curves. Adverse outcomes included the composite outcomes of any adverse events, major adverse events, minor adverse events, and infectious adverse events as well as adverse hospital metrics including discharge to a higher level of care and extended length of hospital stay. (2) For Objective 2, patients undergoing anterior cervical discectomy and fusion procedures were identified in NSQIP and missing preoperative albumin and hematocrit values were handled using complete case analysis and multiple imputation. These preoperative laboratory levels were then tested for associations with 30-day postoperative outcomes using logistic regression. Adverse outcomes included the composite groupings of any adverse event and major adverse events as well as the adverse hospital metric of hospital readmission. Findings were as follows. (1) For Objective 1, although ASA and mFI showed statistically equivalent discriminative ability for the occurrence of any adverse events, major adverse events, minor adverse events, infectious adverse events, and discharge to higher level of care, ASA was statistically better than mFI in identifying patients at high risk for extended length of stay. ASA demonstrated statistically superior discriminative ability for the occurrence of all studied adverse outcomes in comparison with mCCI. (2) For Objective 2, when utilizing complete case analysis, only 4,311 patients of the 11,999 patients in the cohort were studied. The removed patients were significantly younger, healthier, of a numerically common BMI (25-29 kg/m2 and 30-34 kg/m2), and male. Logistic regression failed to identify either preoperative hypoalbuminemia or preoperative anemia as significantly associated with adverse outcomes. When employing multiple imputation, all 11,999 patients were included. Preoperative hypoalbuminemia was significantly associated with the occurrence of any adverse event and severe adverse events. Preoperative anemia was significantly associated with the occurrence of any adverse event, severe adverse events, and hospital readmission. Conclusions are as follows. (1) For Objective 1, ASA, an easily obtained patient comorbidity measure that is based upon the anesthesiologist\u27s evaluation of a patient\u27s physical state prior to surgery, has similar or better discriminative abilities for post-operative adverse outcomes than numerically tabulated indices that have multiple inputs and are harder to calculate. Increased use of this scale should be considered to control for confounding in future spine studies and for risk assessment in the clinical setting. (2) For Objective 2, the use of multiple imputation avoided the loss of cases that can affect the representativeness and power of studies and led to different results than complete case analysis. Future spine studies should consider utilizing multiple imputation, when appropriate