When asked “what is your surgical site infection rate?” one will get an answer that is almost certainly removed from reality.
As a career academic surgeon, it has always been interesting to me how surgeons evaluate complication rates in their own clinical practice.
When asked “what is your surgical site infection rate?” one will get an answer that is almost certainly removed from reality. Is there an intent to deceive? Are they simply making a gross estimate? Do they have selective forgetfulness? Do they even know, since many of their post-operative patients are seen in emergency departments or at hospitals other than where the surgical care was delivered? Objective measurements are lacking for surgical site and other complications of care.
The charade of “What is your surgical site infection rate?” continues when research results are published. Several years ago, I published a study of elective colon surgery from the National Inpatient Sample from the Healthcare Cost and Utilization Project. The surgical site infection rate coded in the discharge abstracts was 3.9% in our study, which I did not believe and stated as such in the manuscript.
For roughly the same time period, the National Healthcare Surveillance Network reported surgical site infection rates in colon surgery between 4-9 % depending upon the risk profile of their study population. The National Surgical Quality Improvement Project reported an overall rate of about 9-11 % in the same procedure. And a prospective, clinical trial performed in elective colon surgery for an FDA approval of the study drug has reported surgical site infections that were greater than 20%.
Clearly different definitions and difference surveillance methods were used. What is the real number and how could any assessment of complication rates be made when there were such different reported rates? What is a complication, anyway?
The Complication of Complication Rates
The cruel reality of 2016 is that we do not know the complication rates of surgical care. For major operations such as colon resections or open-heart surgery, our studies have identified that 40% to 70% of patients will have one or more coded complications in the discharge abstract. Obviously, some are more serious than others and some are inconsequential for patient recovery.
Individual hospitals code very different rates of complications for the same operation and coded complications have no severity indicator. Should a positive urine culture following inpatient urinary tract catheterization that is quickly managed with prompt antibiotic management be given the same equivalency as fulminate postoperative urinary tract sepsis? One will never know from the discharge codes because all are commonly given the same coded designation.
Furthermore, coding selected complications (not urinary tract infection) have the perverse incentive with Medicare Part A hospital payments of increasing revenue for an episode. Despite poor definition of actual complication rates, there is a rush by many patient advocacy groups to publish complication rates by hospital and by clinician in the hopes that this represents discriminating information for the identification of best and worst performance.
What is needed is a single (ie, binary) designation of whether the patient had a major complication or not from his or her inpatient care: that is, a comprehensive and composite measurement of inpatient outcome. To do this, our research group has turned to the writings of Walter Shewhart, the grandfather of manufacturing quality control.
Shewhart was the architect of the concept of Statistical Process Control. He recognized that even with precise industrial manufacturing methods there was very slight variation in the measurements of final widgets that were produced. Common cause variation around the desired specification was acceptable because the final product effectively served its purpose. However, special cause variation was the consequence of failed processes that yielded an end-product that was unacceptable, and if manufacturing processes were producing a defective product, it needed to be promptly identified and corrected. Defective products were consistently ±3-sigma from the manufacturing specification. Can we use statistical process control for the development of a composite measure of inpatient outcomes in hospital care?
Inpatient length-of-stay has become the focus of our quest to apply statistical process control for the composite measurement of severe complications of care. The duration of hospital stay is critically important to hospitals and to insurers. Patients are expected to be discharged in a standard period of time unless an untoward event has occurred. Surgeons recognize this as well, and prompt discharge of post-operative patients with an uneventful recovery is expected. Patients stay for longer periods of time when things have not gone as expected. Thus, we developed linear prediction models from national databases (eg, Medicare data) to define the expected length-of-stay for elective surgical cases when no coded complications were identified, and then apply that prediction model to all cases to identify prolonged length of stay (prLOS) outliers.
Our statistical process control method begins by taking all operative cases of a given type (eg, coronary artery bypass grafting) and we arrange them in temporal sequence of when they were performed during the time period of study. The temporal sequencing in important since statistical process control has time as a dimension in addition to normative statistical distribution. For each hospital being evaluated, the total observed days of hospitalization and the total predicted days are set equal to each other by multiplying the predicted value by the observed:predicted ratio. This adjustment accounts for the local culture of hospitalization practice but maintains the relative weight of the prediction variables. A control chart is then designed and 3-sigma outliers are identified.
Studying Length-of-Stay Outliers
We have found that length-of-stay outliers have been a consistent method for the identification of severe complications of inpatient care. Length-of-stay outliers have significantly higher costs than those patients with coded complications of care but are not outliers. Length-of-stay outliers have highly significant odds ratios in the prediction of 90-day post-discharge deaths and readmissions to the hospital. Even those patients that are length-of-stay outliers but are without coded complications at discharge have higher costs, predict 90-day post-discharge deaths and readmissions, and upon audit consistently have complications that have escaped being coded and have resulted in protracted hospitalization.
We have used this metric in the risk-adjusted evaluation of hospital performance in cardiac surgery that was recently published in The American Journal of Accountable Care.
The use of prolonged length-of-stay outliers has laid a strong foundation for the definition of inpatient complications of surgical care. Occasional cases have been rapidly discharged to skilled nursing facilities when major adverse events have occurred and have escaped identification by prolonged hospitalization. There are uncommon cases where poor discharge planning prior to elective surgical procedures may lead to prolonged hospitalization because of disposition issues and constitute economic morbidity, but not a complication of care. For the vast majority of cases, we have found 3-sigma length-of-stay outliers to be an effective composite measure of inpatient adverse outcomes.
This metric can serve as an effective method to monitor and improve results of surgical care in much the same way that Shewhart envisioned statistical process control over 80 years ago.