HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Add to Personal Folders Download to citation manager Author home page(s): Paul T. Sergeant Eugene H. Blackstone Permission Requests Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Sergeant, P. T. Articles by Blackstone, E. H. Search for Related Content PubMed PubMed Citation Articles by Sergeant, P. T. Articles by Blackstone, E. H.

Ann Thorac Surg 1999;68:362-366
© 1999 The Society of Thoracic Surgeons

---

Data Base Panel

Closing the loop: optimizing physicians’ operational and strategic behavior

^a Katholieke Universiteit Leuven, Leuven, Belgium
^b The Cleveland Clinic Foundation, Cleveland, Ohio, USA

Address reprint requests to Dr Sergeant, Department of Cardiac Surgery, Gasthuisberg University Hospital, Herestraat, 3000 Leuven, Belgium
e-mail: paul.sergeant{at}uz.kuleuven.ac.be

Presented at the Thirty-fourth Annual Meeting of The Society of Thoracic Surgeons, New Orleans, LA, Jan 26–28, 1998.

Abstract

Clinical databases are essential elements in optimizing medical care. They are no finality by themselves, but essential elements in the generation of knowledge. Optimal medical care starts with optimal care based on existing knowledge. This care continues with the registration of the variability in morbidity, comorbidity, and therapy, but also the registration of the early and late outcome. This should then allow the generation of structured inferences based on this registration and the closure of the loop, by treating patients according to this newly created evidence.

Evidence-based operational and strategic behavior of clinicians should be based on knowledge generated by following patients during their treatment, but also beyond their hospital stay, and preferably for as long as possible, until the end of the patient’s life. This knowledge should include the interaction of the patient’s morbidity and comorbidity, the specificity of the disease and the intervention, and environmental factors. The purpose of this manuscript and presentation is to follow the path of these data, from the generation of information up to the generation of knowledge useful for an individual patient or a population.

The generation of information has to follow methodological rules and constraints. The first of these is a methodology for overcoming the absence in the late 20th century of complete, longitudinal, patient records of all health care information. Important morbid events may be forgotten if a patient is contacted at infrequent intervals, or if only vital status information is available from national databases. Therefore, when the K.U. Leuven Coronary Surgery Database was established, a double-core data system for patient follow-up was incorporated.

First, there was an unformalized update by a regular stream of follow-up reports by referring cardiologists. Because most of them trained at our University Hospital, we have encouraged them to send reports after each regular visit. Visits to the cardiologists are financially neutral for the patient in the Belgian Health Care system. In addition to this real-time update, two (1987 and 1993) cross-sectional follow-up studies were performed of known surviving patients using the very strict common closing date methodology. This methodology fulfills the theoretical assumptions of actuarial methods. Others have implemented similar multitiered follow-up systems, but using an anniversary follow-up scheme for obtaining systematic information. This scheme also fulfills the theoretical assumptions of actuarial methods.

This cross-sectional follow-up is managed and monitored by a self-developed information management system, directly integrated in the K.U. Leuven Coronary Surgery Database. Figure 1 describes the different loops of step 1, taking place after defining the common closing date. Figure 2 describes some of the different loops of steps 2 and 3, when the patient is identified as deceased or if events have taken place in follow-up. The loops of step 2 stop only when the senior researcher (a cardiac surgeon) has spoken with a person who was present at the time and place of death of the patient. This is very important for a correct classification of the mode of death, just as important as a complete insight into the events and status of the patient in the weeks and months before death. Death is always sudden and unexpected for the close relatives of a patient. A more profound analysis frequently corrects this information.

View larger version (28K): [in this window] [in a new window]   Fig 1. Step 1 in the generation of information: obtaining the status of the patient at the common closing date.

View larger version (23K): [in this window] [in a new window]   Fig 2. Step 2 (a deceased patient) and step 3 (an event, symptom, or procedure) in the generation of information, once the status of the patient has been identified.

Generally, systems optimized for data gathering, information management, and data updating and verification are not optimized for sophisticated statistical data analysis. Generally, a "snapshot" is needed of the data relevant to the analysis. Importantly, all data corrections discovered during analysis are fed back to the primary database. In our case, data were transferred from the database computing system into the data analysis computing system of the University of Alabama at Birmingham. Even in an "ideal" future setting of existing computerized longitudinal patient medical records, a data analysis phase would require extraction of values for variables across multiple patient records at some specific point in time.

Data analysis must also be approached purposefully, methodically, and with clear questions in mind. The analysis begins with assessing the quality, completeness, and relevance of the study endpoints (outcomes) and variables. These are then organized into medically meaningful clusters. Formalization of this step has become popularly known as an important aspect of "data mining." This is followed by investigation of the relation of each variable in isolation to outcomes. For ordinal (ordered categorical information) and continuous variables, transformations of scale must be sought so as to best fulfill the assumptions of the underlying statistical models.

Table 1 lists some of the steps in data exploration preceding and finally including the generation of patient-specific predictions and nomograms usable for optimizing therapeutic decision making. This process is labor intensive and requires a close and permanent collaboration between the data analyst and the clinician. Only then are the data ready for initial multivariable analysis. The creation of interaction variables and the grouping of categorical variables should be based on extensive clinical experience and previous literature review.

The generation of knowledge starts when the parametric multivariable model has been created [1]. One reason for use of parametric modeling is that the model can be explored graphically (which we have termed "nomograms") to discover the shape of relations, the impact of risk factors on one another, and the appearance and disappearance of evident differences, all of which go into drawing helpful clinical inferences, suggesting hypotheses needing basic research, and developing strategies to optimize therapy in the interim.

This laborious, costly, time-consuming activity of generating new knowledge contrasts with the present intensive scrutiny of in-hospital outcomes. Governments, societies, institutions, and the media use hospital mortality as an indicator of quality of care in cardiac surgery. The information is based on aggregated data for a surgeon or for a center, and most frequently for all surgical interventions combined, uncorrected for type of intervention, pathology, or comorbidity.

It is assumed by many that the operative (30-day) mortality for primary coronary artery bypass grafting (CABG) and for aggregated data should be 2%. Very few authors have actually complete follow-up, even to the 30-day limit; this creates an evident bias in all data without follow-up beyond hospital stay. Figure 3 is the decomposition of the different phases of the hazard of mortality after CABG [2]. The hazard of dying after CABG induced by the preprocedural condition and the procedure stays active for 2 months beyond the data of the procedure. There is no scientific basis to restrict the study of the periprocedural mortality to the hospital stay. Hazard-phase analysis for a particular event identifies the appropriate interval for the study of the periprocedural hazard for that event, for that particular pathology and intervention.

View larger version (19K): [in this window] [in a new window]   Fig 3. Decomposition of the three hazard phases (solid lines) composing the overall instantaneous hazard (dotted line) for mortality by any mode at any time after CABG.

A patient-specific prediction uses a generated equation and uses the patient-specific values for every variable in the equation predicting that particular event. Figure 4 presents a patient specific prediction of the 30-day mortality after CABG, calculated for 9600 consecutive patients (K.U. Leuven 1971–1992) and for the last 1000 patients of that same data set, sorted by increasing mortality. The information presented in this plot is comprehensive.

View larger version (25K): [in this window] [in a new window]   Fig 4. Patient-specific prediction of the 30-day mortality after CABG, calculated for 9600 consecutive patients (K.U. Leuven 1971–1992) and for the last 1000 patients of that same data set, sorted by increasing mortality.

Any statement concerning an average 30-day mortality is therefore groundless. Any national or institutional audit or quality control methodology without extensive correction for this variability is a scientifically invalid method.

Recently, Osswald and colleagues have confirmed that it is the highest risk patients whose risk is most likely to be underestimated by 30-day mortality [3]. However, hospital information is readily available, relatively inexpensive to gather, particularly if confined to administrative and computerized laboratory data, and thus appealing to institutions and providers who tend to focus on short-term issues rather than the long-term care of patients.

Another inference of Figure 4 concerns least invasive techniques of CABG. Aggregated data resulting in a 2% 30-day mortality might seem comparable with standard techniques for the uninformed but are similarly misleading, since the comorbidity of the patients undergoing less-invasive CABG is usually extremely low. The comparable 30-day mortality should be more in the 0.2% to 0.5% range. Randomized trials comparing less-invasive CABG techniques in low-risk patients will therefore have very large sample sizes.

Informed consent to a patient is usually based on a medical cost-benefit calculation. The issue of the cost, expressed as the periprocedural mortality, has been addressed already. The benefit, expressed as the long-term survival, is usually based on an intuitive appreciation by the attending physician of the residual survival reserves. Figure 5 represents the wide spectrum of the survival of our CABG population. Similarly as in 30-day mortality, a very good long-term survival is identified in 25% of the patients, a good one in an additional 25%, and an acceptable one in another 25%. The worst survival is accumulated in the last 1% of the population. It should be clear that only patient-specific predictions, based on long-term follow-up studies, can provide the tools for appropriate medical decision making and informed consent. It is also obvious that honesty toward an audited individual or institution requires that audited interval should be as long as possible, thereby auditing the cost simultaneous with the benefit. Individual surgeons might have an increased periprocedural mortality but with a dramatic gain in late survival, more than compensating for the periprocedural risk.

View larger version (21K): [in this window] [in a new window]   Fig 5. Patient-specific prediction of the 20-year survival after CABG, calculated for 9600 consecutive patients (K.U. Leuven 1971–1992), sorted by decreasing survival. Only the lines representing the minimum, the maximum, and the different 1%, 5%, 10%, 25%, 50%, and 75% are defined.

View larger version (32K): [in this window] [in a new window]   Fig 6. Patient-specific nomogram of the 10-year survival after CABG, for a patient with median values for all risk factors (relevant for this event). The "median patient" is presented with an ejection fraction of 40%, with and without chronic atrial fibrillation and with changing 1-second expiratory volumes (expressed as a percent from normal).

View larger version (17K): [in this window] [in a new window]   Fig 7. Patient-specific nomogram of the 1-year and 5-year difference in freedom from angina after primary CABG, for a patient with median values for all risk factors (relevant for this event). The "median patient" is a triple-vessel disease patient presented with changing age at operation.

Closing the loop, from treating patients, studying their early and late follow-up, and using that generated knowledge in the treatment of the future patients, is what medicine should be all about. The generated knowledge can cover many domains. Some of these domains are the variability in pathology, in treatment and in patient spectrum. The generated knowledge should help the attending physician in the actual care of an individual patient by solving the differential therapy problem, a more documented informed consent, optimizing therapeutic strategy at primary care, and at return of symptoms. Evidence-based medicine should similarly guide departmental management as well as departmental quality control.

References

1. Blackstone E., Naftel D., Turner M. The decomposition of time-varying hazards into phases, each incorporating a separate stream of concomitant information. J Am Stat Assoc 1986;81:615-624.
2. Sergeant P., Blackstone E., Meyns B. K.U. Leuven Coronary Surgery Program. Validation and interdependence with patient-variables of the influence of procedural variables on early and late survival after CABG. Eur J Cardiothorac Surg 1997;12:1-19.[Abstract]
3. Osswald B.R., Blackstone E.H., Tochtermann U., Thomas G., Vahl C.F., Hagl S. The meaning of early mortality after CABG. Eur J Cardiothorac Surg 1999;15:401-407.[Abstract/Free Full Text]
4. Kirklin J., Barratt-Boyes B. Patient-specific predictions and comparisons in ischemic heart disease. In: Kirklin J., Barratt-Boyes B., eds. Cardiac surgery II. New York: Churchill Livingstone, 1991:344-368.
5. Sergeant P., Blackstone E., Meyns B. Is return of angina after coronary artery bypass grafting immutable, can it be delayed, and is it important?. J Thorac Cardiovasc Surg 1998;116:440-453.[Abstract/Free Full Text]
6. Sergeant P., Blackstone E., Meyns B. Can the outcome of coronary bypass grafting be predicted reliably?. Eur J Cardiothorac Surg 1997;11:2-9.[Abstract]

This article has been cited by other articles:

				L. Noyez, F. W.A. Verheugt, and H. A. van Swieten The importance of an organized follow-up for the evaluation of mortality after hospital discharge in cardiac surgery Interactive CardioVascular and Thoracic Surgery, June 1, 2008; 7(3): 449 - 451. [Abstract] [Full Text] [PDF]

				W. G. Williams Surgical outcomes in congenital heart disease: expectations and realities Eur. J. Cardiothorac. Surg., June 1, 2005; 27(6): 937 - 944. [Abstract] [Full Text] [PDF]

				A.A. Albert, J.A. Walter, B. Arnrich, W. Hassanein, U.P. Rosendahl, S. Bauer, and J. Ennker On-line variable live-adjusted displays with internal and external risk-adjusted mortalities. A valuable method for benchmarking and early detection of unfavourable trends in cardiac surgery Eur. J. Cardiothorac. Surg., March 1, 2004; 25(3): 312 - 319. [Abstract] [Full Text] [PDF]

				M.-B. Edwards and K. M. Taylor Is 30-day mortality an adequate outcome statistic for patients considering heart valve replacement? Ann. Thorac. Surg., August 1, 2003; 76(2): 482 - 485. [Abstract] [Full Text] [PDF]

				P J Bradshaw, K Jamrozik, M Le, I Gilfillan, and P L Thompson Mortality and recurrent cardiac events after coronary artery bypass graft: long term outcomes in a population study Heart, December 1, 2002; 88(5): 488 - 494. [Abstract] [Full Text] [PDF]

				F. L. Grover The Society of Thoracic Surgeons National Database: current status and future directions Ann. Thorac. Surg., August 1, 1999; 68(2): 367 - 373. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Add to Personal Folders Download to citation manager Author home page(s): Paul T. Sergeant Eugene H. Blackstone Permission Requests Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Sergeant, P. T. Articles by Blackstone, E. H. Search for Related Content PubMed PubMed Citation Articles by Sergeant, P. T. Articles by Blackstone, E. H.