|Home | About | Journals | Submit | Contact Us | Français|
Patients are often treated in hospital by physicians other than their regular community doctor. After they are discharged, their care is often returned to their regular community doctor and patients may not see the hospital physician. Transfer of information between physicians can be poor. We determined whether early postdischarge outcomes changed when patients were seen after discharge by physicians who treated them in the hospital.
This cohort study used population-based administrative databases to follow 938,833 adults from Ontario, Canada, after they were discharged alive from a nonelective medical or surgical hospitalization between April 1, 1995, and March 1, 2000. We determined when patients were seen after discharge by physicians who treated them in the hospital, physicians who treated them 3 months prior to admission (community physicians), and specialists. The outcome of interest was 30-day death or nonelective readmission to hospital.
Of patients studied, 7.7% died or were readmitted. The adjusted relative risk of death or readmission decreased by 5% (95% confidence interval [CI], 4% to 5%) and 3% (95% CI, 2% to 3%) with each additional visit to a hospital physician rather than a community physician or specialist, respectively. The effect of hospital physician visits was cumulative, with the adjusted risk of 30-day death or nonelective readmission reduced to 7.3%, 7.0%, and 6.7% if patients had 1, 2, or 3 visits, respectively, with a hospital rather than a community physician. The effect was consistent across important subgroups.
Patient outcomes could be improved if their early postdischarge visits were with physicians who treated them in hospital rather than with other physicians. Follow-up visits with a hospital physician, rather than another physician, could be a modifiable factor to improve patient outcomes following discharge from hospital.
Hospital readmissions are costly 1 and common.2 Studies have identified factors that increase readmission risk, including patient age,3,4 male gender,4 length of stay,5 complications during hospitalization, comorbidities,4–6 medical rather than surgical admissions,4 particular diagnoses,7–9 increased case complexity,10 low patient socioeconomic status,11 day of discharge,12 and prior hospital utilization.3–5,9,13–15 This research identifies patients with a high risk of readmission. However, these factors are primarily immutable.
Modifiable factors associated with poor outcomes after discharge suggest interventions to improve patient care. Coordination of care is one such factor. Care can be coordinated by improving continuity of information or increasing the continuity of care. The former occurs when patients are treated by physicians who know what other physicians who have been treating the patient are doing. Postdischarge outcomes may improve when patient hospital information is disseminated to follow-up physicians.16 However, transfer of patient information between hospitals and community physicians can be poor.17–23 Therefore, hospital-physician continuity of care—which occurs when patients are seen in follow-up by physicians who treated them during the hospitalization—may be particularly important when patients are discharged from hospital.
One randomized trial tested whether outcomes after discharge changed with increased hospital-physician continuity of care.24 Male veterans discharged from hospital were randomized to regular care or intensive follow-up provided by study physicians and nurses who saw the patient prior to discharge. Over a 6-month period, there was no difference between the intervention and control group in the likelihood of nonelective readmissions (80% vs 77%) or preventable readmissions (35% vs 37%). The generalizability and validity of this trial has been questioned because the patients were likely unrepresentative,25,26 and the intervention health staff did not actively treat the patient during the hospitalization.27
The effect of hospital-physician continuity of care upon early outcomes is increasingly important because family physicians are less likely to treat hospitalized patients.28 We therefore used a population-based cohort study to determine whether outcomes changed when physicians who cared for patients during the hospitalization saw them in follow-up.
This cohort study took place in Ontario, Canada. During the study period, between 50% and 60% of Ontario family physicians cared for inpatients.29 In 1998, 33.8% of medical inpatients were cared for, at least in part, by their family physician.30
This study used 3 administrative databases. The Discharge Abstract Database (DAD) records all admissions to Ontario hospitals and documents demographic, diagnostic, procedural, and hospitalization information in a standardized fashion. The Physician Services Database (PSD) records physician claims for visits of more than 95% of family physicians and almost all specialists. The date of each visit, the patient, and the specialty of the physician are recorded for each claim. The Registered Patients Database (RPD) records the death date of all Ontarians, including those who died outside of the province. All databases are anonymous and were linked by common patient unique identifiers. The study was approved by the research ethics board of the Sunnybrook and Women's College Hospital.
The DAD identified all Ontarians greater than 20 years of age who were admitted and discharged alive from an acute care hospital between April 1, 1995 and March 1, 2000. Only nonelective medical or surgical admissions were included because such patients are substantially different than electively admitted patients. Patients were excluded if they were admitted from or discharged to another hospital, had an invalid patient identifier, lacked key information, or were significant outliers by exceeding the 99th percentile for hospital length of stay, previous hospitalizations, number of follow-up visits, or case resource consumption. Patients were also excluded if they were discharged from one of two hospitals whose physicians do not submit claims to the PSD. Finally, some patients were admitted multiple times during the study. To ensure that all observations in the dataset were truly independent, we randomly selected one admission for each patient.
We identified demographic factors associated with hospital readmission. From the DAD, we determined patient age, gender, comorbidities (measured using the Deyo modification of the Charlson score 31), and each person's postal code. The postal code was linked to 1996 Canadian Census information to determine whether patients lived in an urban or rural area and the median household income of each person's area of residence. The latter was used as a proxy measure of socioeconomic status.
Prehospitalization factors included whether patients had a nonelective hospitalization during the 6 months prior to the index admission. This was determined using the DAD. Using the PSD, we also measured the number of physician visits, and the number of different physicians visited, during the 3 months prior to admission.
Hospitalization factors included the hospital length of stay. The discharge diagnostic category was classified by the Case Mix Group (CMG), determined by a Canadian Institute for Health Information (CIHI) algorithm that groups patients with similar diagnoses using diagnostic and procedural codes in the DAD. The CMG was used to classify admissions as medical or surgical. Other hospital information included the weekday of discharge and whether the patient experienced a procedure or complication during the hospitalization. Complications were determined by the presence of any type-2 International Classification of Diseases, Ninth Revision (ICD-9) diagnoses. Finally, case resource consumption was measured using the Resource Intensity Weight (RIW), another CIHI methodology measuring hospitalization costliness with the DAD.
Follow-up factors were all measured with the PSD. We determined the date of all physician visits in the month after discharge from hospital. To measure the effect of community continuity of care, we determined which postdischarge visits were with physicians who saw the patient in the 3 months prior to hospitalization (herewith called a “community physician”). To measure the effect of specialty, we determined which postdischarge visits were with a “specialist” (physician whose specialty code in the PSD was not “general practitioner” or “family physician”). In Canada, internists function as consultants and generally do little primary care. To measure the effect of hospital continuity of care, we determined which postdischarge visits were with physicians who treated the patients during the hospitalization (referred to as a “hospital physician”). The DAD records hospitalization admit and discharge dates. Physicians with assessment claims between admit and discharge dates were a hospital physician for that patient. Community physician, specialist, and hospital physician visits are not mutually exclusive. For example, a particular visit would be with both a community physician and a hospital physician if the physician saw that patient prior to and during the hospitalization. The multivariate analysis in the study was able to distinguish the independent effect of each follow-up physician type.
The study outcome was death or nonelective readmissions to hospital in the first 30 days. This combined outcome avoids the bias of censoring deaths when hospital readmission alone is examined.32 In addition, the relationship between readmission and survival varies between diagnoses.7 Readmissions were identified in the DAD. Deaths were identified in the RPD.
To determine the effect of follow-up visits upon outcomes, we determined the cumulative number of each visit type for each patient during the first 30 days after discharge. These are time-dependent variables because their values change over time. We therefore used proportional hazards techniques with time-dependent covariates to model their effect upon patient outcome after controlling for significant confounders.33 Time-dependent covariate analysis avoids survivor treatment selection bias.34
To build our models, we identified demographic, prehospitalization, and hospitalization factors that were significantly (two-sided P value of .05) associated with 30-day death or nonelective readmission in a univariate proportional hazards model. These variables were offered to a backward-stepping proportional hazards model that determined which were independently associated with the outcome. These independent variables were included in the final model with follow-up visits types (i.e., cumulative number of all physician visits, community physician visits, specialist visits, and hospital physician visits) expressed as time-dependent covariates. The hazard ratios for contrasting follow-up visit types were calculated from this model (see Appendix A, available at http://www.jgim.org). For example, the hazard ratio for visiting a hospital physician rather than a community physician was e(β[hospital physician visit]−β[community physician visit]).Each model was stratified by hospital, which allows the underlying hazard function to vary between hospitals and adjusts for unmeasured hospital factors.35 We reasoned that patients with more physician visits are sicker and are more likely to die or be readmitted to hospital and controlled for this by including the cumulative number of all physician visits in all models.
Readmission risk varies widely between diagnoses.8 To control for the effect of diagnosis upon outcome, we used established methodology 36 to calculate risk scores for each diagnostic group (Appendix B), available at http://www.jgim.org).37
We conducted several subanalyses to measure the consistency of our findings. Analysis was limited to subgroups of the most predictive baseline factors including propensity score quintiles. We repeated the analysis for patients whose number of physician visits after discharge did not exceed the 75th percentile (3 visits) to determine whether outlier patients with many follow-up visits influenced the association. We a priori identified common diagnoses whose postdischarge course either could be, or should not be, influenced by hospital physician follow-up (Appendix C), available at http://www.jgim.org). We hypothesized that hospital physician follow-up should be more beneficial in the former diagnoses. Finally, similar to previous studies,38 we conducted the analysis separately for the 100 most common discharge diagnoses and used the Sign test 39 to determine whether the number of diagnoses in which hospital physician follow-up was protective exceeded that expected by chance.
During the study period, 2,607,416 adults had a nonelective hospitalization. Some 1,668,583 hospitalizations were excluded because the patient died during the hospitalization (n= 89,756), the patient was transferred to or from another hospital (n= 97,885), the patient had an invalid Ontario Health Insurance Plan (OHIP) number (n= 5), the patient was discharged from an excluded hospital (n= 19,287), the patient was an outlier (n= 14,614), the patient was missing data (n= 19,246), or the hospitalization was randomly excluded because the patient had other hospitalizations during the study period (n= 1,427,790).
This left 938,833 patients in the study (Table 1) Most patients had Charlson-Deyo scores of zero. The most common medical diagnostic groups included acute gastroenteritis or gastrointestinal bleeding (8.6% of medical diagnoses), pneumonia (5.1%), and chest pain (5.1%). For surgery, the most common groups included acute fractures (8.8%), appendectomy (7.3%), and laparoscopic cholecystectomy (6.1%). Patients were spread equally through the 5 study years and came from 185 different hospitals. During the entire study, 43.5% of patients were treated in hospital by a physician who had seen them during the 3 months prior to admission. This proportion decreased steadily through the study from 45.6% in 1995 to 41.0% in 1999.
Overall, 71,944 (7.7%) patients had an event (Table 1). Deaths accounted for 14.6% of events. The risk of death or readmission was highest if patients were older, had been hospitalized in the last 6 months, had a diagnostic group with a high-risk score, or had a medical admission (Table 1). The relative risk of death or readmission decreased 3% with each year beyond 1995 (independent hazard ratio, 0.97; 95% confidence interval [CI], 0.96 to 0.98). The univariate analysis showed that the risk of death or readmission increased with each additional visit with all physicians, with community physicians, with specialists, or with hospital physicians (Table 1).
Patients had a median of 2 physician visits in the first month after discharge (Table 1). Approximately half of these were with hospital physicians. Of the patients, 751,775 (80.1%) had one or more physician visits, 473,814 (50.5%) saw one or more community physicians, 448,035 (47.7%) saw one or more specialists, and 662,029 (70.5%) saw one or more hospital physicians. Patients who saw hospital physicians appeared sicker because they were older, had higher Charlson and diagnostic risk scores, and were more likely to have a medical admission (Table 2) This was true even when patients with no follow-up visits were excluded (Table 2).
After controlling for important confounders, patients were significantly less likely to die or be readmitted if they were seen in follow-up by a hospital physician rather than a community physician (hazard ratio, 0.95; 95% CI, 0.95 to 0.96) or specialist (hazard ratio, 0.97; 95% CI, 0.97 to 0.98;Table 3) This means that the relative risk of death or readmission decreased by 5% (95% CI, 2% to 4%) when patients followed up with a hospital rather than a community physician. Given a baseline risk of 7.7%, the adjusted risk of 30-day death or nonelective readmission would be 7.3%, 7.0%, and 6.6% for patients who had 1, 2, or 3 visits, respectively, with a hospital rather than a community physician.
The protective effect of hospital physician follow-up was consistent within important subgroups with some possible exceptions (Table 4) Compared to community physicians, hospital physician follow-up was significantly protective for all subgroups except patients with previous hospitalizations. Compared to specialists, hospital physician follow-up was significantly protective for all subgroups except those with previous hospitalizations, elderly patients, and those with high-risk diagnoses. The benefit of hospital physician follow-up appeared to decrease as the diagnostic risk score increased. Protection with hospital physician follow-up remained when the analysis was limited to patients whose total number of postdischarge visits did not exceed the 75th percentile. Hospital physician follow-up was more effective for diagnoses thought likely be sensitive to hospital physician follow-up (Table 4);Appendix C, online). Follow-up with a hospital physician versus a community physician was equally beneficial when analysis was limited to patients who were not seen in hospital by community physicians (hazard ratio, 0.95; 95% CI, 0.95 to 0.96). Finally, follow-up with a hospital physician appeared protective (i.e., had a hazard ratio of less than 1.0) for 70 out of the 100 most common diagnoses, which significantly exceeds that expected by chance (P < .0001).
Using population-based administrative databases over a 5-year period, we found that patients having hospital physician follow-up were significantly and independently less likely to die or get urgently readmitted to hospital in the first 30 days following discharge. After controlling for other important factors, the risk of death or readmission decreased when patients were seen by a hospital physician rather than by another physician. This association was consistent across important subgroups.
Although the effect of hospital physician follow-up was smaller than other factors, our findings are still important. Nonelective readmission and death are both clearly important outcomes and any decrease in their frequency is desirable. These events affect many people, thereby making even small decreases in their frequency significant. The effect of hospital physician follow-up is cumulative so that the risk of death or readmission decreases with each visit. Most importantly, hospital physician follow-up is a potentially modifiable factor that could decrease the risk of bad outcomes post discharge.
In contrast to continuity of care in the community,40–43 the effect of continuity of care with hospital physicians has not been studied extensively. Several issues explain why our results differ from those of Weinberger et al.24 We only examined the first 30 days after discharge from hospital. Our study included all Ontario hospitals and a diverse patient population. Also, all of the hospital physicians in our study treated the patients during, and commonly throughout, their hospitalization. Because our study was population based, included readmissions to all hospitals, and adjusted for important confounders, we believe that it is a very representative assessment of the effect that continuity of care after discharge from hospital has upon patient outcomes.
Hospital physician follow-up could improve outcomes through several mechanisms. Hospital information is often inadequately transferred to community physicians 17 and is key to evaluating a patient in the early postdischarge time period. Familiarity with the hospital course allows follow-up physicians to determine therapeutic effectiveness and identify complications of hospital therapies or procedures. Complications that are dealt with early could avoid more serious subsequent problems. Finally, patients are often discharged from hospital with problems that are improving but not yet resolved. Therefore, patients seen early following discharge from hospital can still be very ill. If a physician does not know that a particular patient was worse when they were admitted to hospital, their condition could be interpreted as a deterioration requiring readmission.
This last point highlights a limitation of our study. Although we measured urgent readmissions to all Ontario hospitals, we were unable to determine the appropriateness of these readmissions. As with any health service, physician practice patterns can influence the decision to admit a patient, possibly independent of a patient's clinical status. Other factors extraneous to the patient's health can also modify the decision to admit a patient to hospital. Despite the “noise” that these factors introduce to this outcome, we still found an independent and significant association between hospital physician follow-up and improved outcomes.
Several aspects of our data support a cause-and-effect relationship between hospital physician follow-up and improved outcomes.44 There are many reasons to expect better outcomes with hospital physician follow-up. We found a dose-response effect with further improvement of outcomes with each additional hospital physician visit. Our association was consistent in several distinct populations. Finally, there is no ambiguity in the timing of the hospital physician visit in relation to the death or readmission.
However, our study does not absolutely establish better outcomes with hospital physician follow-up. Relative to other factors, the independent association of hospital physician follow-up is small. Hospital physician follow-up was not enormously more successful in diagnoses that we thought would be more sensitive to such visits, although this classification is admittedly rather crude and very susceptible to exceptions. This could be from our inability to control for factors that might contribute to outcomes, such as hospital quality of care.45–48 Our study also did not measure or control for dissemination of patient-specific information after discharge from hospital, which may influence patient outcomes.16 Finally, although the analysis adjusted for many important factors that could influence patient outcome and the results were very consistent in many pertinent subgroups, there still remains a possibility that patients who are seen in follow-up by hospital physicians are systematically distinct from those who are followed by community physicians in important factors that are not measured in the databases used in this study. A better understanding of the effect that hospital physician follow-up has upon patient outcomes requires further research.
What should physicians, patients, and administrators do until such evidence is available? We believe that patients who require physician assessment after being discharged from hospital should be seen by physicians who actually cared for them during the hospitalization. At the very least, physicians who see the patient should have access to as much information regarding the hospitalization as possible.
We thank Drs. Alan Forster, David Juurlink, Alan Karovitch, Stephen Kravcik, Andreas Laupacis, Graham Nichol, Jim Nishikawa, Donald Redelmeier, Tom Stelfox, and James Watters for comments on initial drafts of this paper.
Dr. van Walraven is an Ontario Ministry of Health Career Scientist.
Dr. van Walraven conceived the study and is the responsible author. All authors contributed to the study's design, analysis, interpretation, and writing.