Joint replacement is a costly and frequently performed inpatient procedure. In 1995, Henry Ford Hospital carried out 280 total hip replacements and 225 total knee replacements. To decrease unnecessary inpatient utilization and enhance functional outcomes of joint replacement surgery, Henry Ford Hospital established a multi-disciplinary improvement team, including representatives from Orthopedic Surgery, Physical Therapy, Home Health Care, Social Work, Utilization Management, the Center for Clinical Effectiveness, and Marketing.
The team developed and implemented a best practice guideline in the form of a “care-map”, describing the default plan of care for delivering multi-disciplinary services to the patient for each day of the planned length of stay. These interventions included outcomes assessments, tests & diagnostics, consults, treatments, procedures, medications, diet & nutrition, elimination goals, activity goals, and safety goals, skin condition goals, educational interventions and learning goals, and discharge planning.
To support the evaluation of the work of the team, comparative data was used. These data included comparative outcomes data obtained through a collaborative Outcomes Measurement Consortium organized through the American Medical Group Association (AMGA). Comparative process data was obtained through the Group Practice Improvement Network (GPIN). In addition, a periodic audit process was used to measure variance from the care-map. Additional process and outcomes data were collected and managed using software applications developed by the Center for Clinical Effectiveness:
- Complications Tracking System used to enter and report on trends in various joint-replacement complications. This system was used to support the Department of Orthopedic Surgery morbidity and mortality conferences.
- Outcomes data acquisition was accomplished using the “Flexi-Scan” forms scanning and study management software.
- Cross-institutional pooling of outcomes data data pooling and quarterly analysis was accomplished by staff of the American Group Practice Association usuing the Flexi-Scan analytic dataset builder and other tools developed by the Center for Clinical Effectiveness.
Finally, patient satisfaction and subjective feedback data was obtained using patient focus groups arranged by Center for Clinical Effectiveness and staff from the Marketing Department.
The implementation of the care-map led to a further one-day reduction of the length-of-stay, as illustrated in the following figure.
Functional outcomes data revealed that hip replacement surgery led to rapid reduction in bodily pain to normal age and sex-adjusted levels. (Note that in the following graph, pain is expressed on the SF-36 pain scale, in which higher numbers represent better functional status, or less pain).
As shown in the following figure, physical function is also improved, although not as rapidly nor as dramatically as bodily pain.
The following figure shows the distribution in bodily pain and physical function, showing that although the average improvement is favorable, 15% of patients have worse pain one year after surgery, and 22% have worse physical function after surgery.
An analysis was conducted to identify baseline variables (collected prior to the surgery) which could predict failure to acheive improvement in pain and function. As shown in the following table, mild pre-operative pain was among the strongest predictors of failure to achieve an improvement in pain.
In order to optimize the ability to predict which patients would fail to achieve a pain reduction from hip replacement, a neural network was trained based on 13 baseline variables collected from the patient before surgery. The neural network was then used to calculate a predictive score for each of the patients. The frequency distribution of predictive scores for patients that did achieve a pain benefit, as well as the distribution for those that did not experience a pain reduction are shown in the following figure.
If the neural network was perfectly predictive, these two distributions would not overlap at all. A threshold predictive score was selected such that the model could be said to be 85% sure about making a prediction that a given patient would not achieve a pain reduction from surgery. Of the 185 patients that experieced an improvement, only 2 had a neural network predictive score below 0.3. Of the 36 patients that experienced an improvement, 13 of them had scores under 0.3. So, for the 15 patients with a score below 0.3, 85% failed to experience a pain reduction. The sensitivity of the model in predicting this type of treatment failure was 35%. In other words, of 100 patients considering hip replacement therapy, about 15 of these will fail to achieve a pain reduction. Of the 15, about 5 of these patients can be identified in advance using this neural network and using a cut-off threshold of 0.3. If implemented in clinical practice, this model could lead to the elimination of the treatment cost and risk of complications associated with these 5 patients. On the down side, approximately one patient who would have benefited from the surgery would have been incorrectly told that they were unlikely to benefit. Such a patient would unnecessarily suffer pain that could have been relieved by the surgery.
As shown in the following figure, analysis also revealed that patients had unrealistic expectations of the time it would take for them to feel fully recovered after the surgery.
Finally, the data revealed that, compared to other institutions, Henry Ford experienced an unusually long physical function recovery period among patient receiving revision total hip replacement (operations on patients that had previously received hip replacement surgery). Such data, shown in the following figure, led to a re-evaluation of Henry Ford’s approach to post-operative rehabilitation.