Comparing Methods to Identify General Internal Medicine Clinic
Objectives. Identification of patients with left ventricular systolic dysfunction is the first step in identifying which patients may benefit from clinical practice guidelines. The purpose of this study was to develop and validate a computerized tool using clinical information that is commonly available to identify patients with left ventricular systolic dysfunction (LVSD).
Methods. We performed a cross-sectional study of patients seen in a Department of Veterans Affairs General Internal Medicine Clinic who had echocardiography or radionuclide ventriculography performed as part of their clinical care.
Results. We identified 2246 subjects who had at least one cardiac imaging study. A total of 778 (34.6%) subjects met study criteria for LVSD. Subjects with LVSD were slightly older than subjects without LVSD (70 years vs 68 years, P = .00002) but were similar with regard to sex and race. Subjects with LVSD were more likely to have prescriptions for angiotensin-converting enzyme (ACE) inhibitors, carvedilol, digoxin, loop diuretics, hydralazine, nitrates, and angiotensin II receptor antagonists. Of the variables included in the final predictive model, ACE inhibitors, loop diuretics, and digoxin exerted the greatest predictive power. Discriminant analysis demonstrated that models containing pharmacy information were consistently more accurate (75% accurate [65% sensitivity, 81% specificity]) than those models that contained only International Classification of Diseases, 9th revision (ICD-9), codes, including ICD-9 codes for congestive heart failure (72% accurate [80% sensitivity, 68% specificity]).
Conclusions. We demonstrated that an automated, computer-driven algorithm identifying LVSD permits simple, rapid, and timely identification of patients with congestive heart failure by use of only routinely collected data. Future research is needed to develop accurate electronic identification of heart failure and other common chronic conditions.
A myriad of clinical practice guidelines have been developed in an effort to improve the quality of care for numerous conditions. These guidelines typically recommend therapies that have been proved to be beneficial and are organized to consolidate expert opinions into usable form. In principle, by implementing guidelines into clinical practice, variability can be reduced and outcomes improved. However, given the tremendous proliferation of current recommendations in guidelines, it is increasingly apparent that the implementation of most guidelines will need to be automated as much as possible. Research has shown that computerized reminders of guidelines are one of the most effective ways to change provider behavior, particularly prescribing habits.
In creating guidelines, a critical initial step that is often ignored is ascertaining how to identify patients to whom the guidelines pertain. Somehow, developers often blithely assume that any health care organization that elects to implement its guidelines will possess an information system capable of easily enumerating all the patients to whom the guideline would be applicable. Unfortunately, in practice this is rarely the case. It is important to ensure that any selection criteria applied possess high levels of both sensitivity and specificity. Possible means of identifying eligible patients include querying clinical and administrative databases for discharge diagnoses or other relevant data. Reviewing medical records is a less attractive alternative because it is inefficient and expensive.
Guidelines for the treatment of chronic heart failure (CHF), for example, have been widely advocated to promote better outcomes. An extensive body of literature has demonstrated that several pharmacologic therapies such as angiotensin-converting enzyme (ACE) inhibitors, b-blockers, and spironolactone can significantly decrease mortality and may improve symptoms of CHF. Although guidelines for management of CHF exist, case finding is difficult. Signs and symptoms of CHF (eg, dyspnea, fatigue, and edema) are nonspecific and may be subtle in many patients. Imaging studies, although sensitive and specific, are not always obtained to investigate these symptoms. Patients with controlled CHF are often treated on an outpatient basis, making hospital discharge diagnoses an insensitive indicator. In addition, discharge diagnoses may be inaccurate because of inadvertent or intentional coding errors. On the other hand, outpatient diagnoses are often incomplete. Moreover, current treatment guidelines for CHF do not apply to all patients with CHF. Rather, they apply only to those with left ventricular systolic dysfunction (LVSD). Medical management for left ventricular diastolic dysfunction is considerably different. As many as 40% of patients with heart failure have preserved systolic function, excluding them from published guidelines developed for patients with left ventricular dysfunction. Because current International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes for heart failure do not distinguish between left ventricular diastolic dysfunction and LVSD, use of discharge diagnoses alone will identify many patients for whom current guidelines are not relevant. Currently, only a cardiac imaging test such as echocardiography or radionuclide studies can accurately differentiate between these 2 groups of patients.
In this study we sought to compare different methods for using commonly available computerized data to identify patients with LVSD who would be candidates for treatment according to currently accepted treatment guidelines.
Objectives. Identification of patients with left ventricular systolic dysfunction is the first step in identifying which patients may benefit from clinical practice guidelines. The purpose of this study was to develop and validate a computerized tool using clinical information that is commonly available to identify patients with left ventricular systolic dysfunction (LVSD).
Methods. We performed a cross-sectional study of patients seen in a Department of Veterans Affairs General Internal Medicine Clinic who had echocardiography or radionuclide ventriculography performed as part of their clinical care.
Results. We identified 2246 subjects who had at least one cardiac imaging study. A total of 778 (34.6%) subjects met study criteria for LVSD. Subjects with LVSD were slightly older than subjects without LVSD (70 years vs 68 years, P = .00002) but were similar with regard to sex and race. Subjects with LVSD were more likely to have prescriptions for angiotensin-converting enzyme (ACE) inhibitors, carvedilol, digoxin, loop diuretics, hydralazine, nitrates, and angiotensin II receptor antagonists. Of the variables included in the final predictive model, ACE inhibitors, loop diuretics, and digoxin exerted the greatest predictive power. Discriminant analysis demonstrated that models containing pharmacy information were consistently more accurate (75% accurate [65% sensitivity, 81% specificity]) than those models that contained only International Classification of Diseases, 9th revision (ICD-9), codes, including ICD-9 codes for congestive heart failure (72% accurate [80% sensitivity, 68% specificity]).
Conclusions. We demonstrated that an automated, computer-driven algorithm identifying LVSD permits simple, rapid, and timely identification of patients with congestive heart failure by use of only routinely collected data. Future research is needed to develop accurate electronic identification of heart failure and other common chronic conditions.
A myriad of clinical practice guidelines have been developed in an effort to improve the quality of care for numerous conditions. These guidelines typically recommend therapies that have been proved to be beneficial and are organized to consolidate expert opinions into usable form. In principle, by implementing guidelines into clinical practice, variability can be reduced and outcomes improved. However, given the tremendous proliferation of current recommendations in guidelines, it is increasingly apparent that the implementation of most guidelines will need to be automated as much as possible. Research has shown that computerized reminders of guidelines are one of the most effective ways to change provider behavior, particularly prescribing habits.
In creating guidelines, a critical initial step that is often ignored is ascertaining how to identify patients to whom the guidelines pertain. Somehow, developers often blithely assume that any health care organization that elects to implement its guidelines will possess an information system capable of easily enumerating all the patients to whom the guideline would be applicable. Unfortunately, in practice this is rarely the case. It is important to ensure that any selection criteria applied possess high levels of both sensitivity and specificity. Possible means of identifying eligible patients include querying clinical and administrative databases for discharge diagnoses or other relevant data. Reviewing medical records is a less attractive alternative because it is inefficient and expensive.
Guidelines for the treatment of chronic heart failure (CHF), for example, have been widely advocated to promote better outcomes. An extensive body of literature has demonstrated that several pharmacologic therapies such as angiotensin-converting enzyme (ACE) inhibitors, b-blockers, and spironolactone can significantly decrease mortality and may improve symptoms of CHF. Although guidelines for management of CHF exist, case finding is difficult. Signs and symptoms of CHF (eg, dyspnea, fatigue, and edema) are nonspecific and may be subtle in many patients. Imaging studies, although sensitive and specific, are not always obtained to investigate these symptoms. Patients with controlled CHF are often treated on an outpatient basis, making hospital discharge diagnoses an insensitive indicator. In addition, discharge diagnoses may be inaccurate because of inadvertent or intentional coding errors. On the other hand, outpatient diagnoses are often incomplete. Moreover, current treatment guidelines for CHF do not apply to all patients with CHF. Rather, they apply only to those with left ventricular systolic dysfunction (LVSD). Medical management for left ventricular diastolic dysfunction is considerably different. As many as 40% of patients with heart failure have preserved systolic function, excluding them from published guidelines developed for patients with left ventricular dysfunction. Because current International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes for heart failure do not distinguish between left ventricular diastolic dysfunction and LVSD, use of discharge diagnoses alone will identify many patients for whom current guidelines are not relevant. Currently, only a cardiac imaging test such as echocardiography or radionuclide studies can accurately differentiate between these 2 groups of patients.
In this study we sought to compare different methods for using commonly available computerized data to identify patients with LVSD who would be candidates for treatment according to currently accepted treatment guidelines.