Abstract
The association of mortality with patient factors (severity of illness, comorbidity), physician factors (specialty training, prehospitalisation visit, in-hospital consultation, volume of patients seen per physician) and healthcare organisation factors (patient-travel distances, regional beds per capita, admitting hospital-bed occupancy, admitting hospital-bed turnover, hospital location, volume of pneumonia cases per hospital) after hospital admission with community-acquired pneumonia was investigated using administrative data from Alberta, Canada from April 1, 1994–March 31, 1999.
During the 5‐yr study period there were 43,642 pneumonia hospitalisations, with an 11% in-hospital and 26% 1‐yr mortality. Patient severity of illness and comorbidity were the strongest predictors of increased mortality. Physicians with the highest in-hospital pneumonia patient volume (>27 patients·yr−1) cared for patients with greater severity/comorbidity, but with decreased odds of in-hospital mortality, compared with the lowest volume physicians (less than seven patients per year).
The effects of internal medicine specialist or subspecialist care were mixed, with a reduction in deaths for the first 72 h and an increase in in-hospital deaths. Pre-hospitalisation visit by a physician was associated with decreased mortality. Healthcare organisation factors were the least strong predictor of mortality, demonstrating an effect only for 1‐yr mortality in those discharged alive from hospital. Admissions to larger volume or metropolitan hospitals were associated with a decrease in mortality.
Severity of illness and comorbidity had the strongest association with mortality. The first association of high-volume physician and pre-hospital care with decreased in-hospital mortality for community-acquired pneumonia is reported.
This work was partially supported by the Alberta Center for Health Service Utilisation Research.
Community-acquired pneumonia (CAP) is a common reason for hospital admission 1, 2. Algorithms to help clinicians decide upon hospital admission and discharge have been devised, validated and published 2–6. These guidelines are robust and may decrease mortality 7, 8. Despite clinical guidelines, considerable variation exists in the treatment of CAP within larger Canadian hospitals 9. The Healthcare Financing Administration stated that peer-reviewed organisations also documented variations in the processes of care for hospitalised CAP 10 that directly impacted upon patient outcome 11, 12. All these studies address quality-of-care issues at the level of the clinician-patient interaction. There are few studies of healthcare system factors that are associated with different mortality outcomes in hospitalised pneumonia patients.
This study attempts to quantify the relative strength of associations between mortality for hospitalised CAP and patient factors (severity of illness, comorbidity), physician factors (specialty training, pre-hospitalisation visit, in-hospital consultation, volume of patients seen per physician), and healthcare organisation factors (patient travel distances, regional beds per capita, admitting hospital-bed occupancy, admitting hospital-bed turnover, hospital location, volume of pneumonia cases per hospital, season/day/time of admission). The association of all these factors to in-hospital mortality and 1‐yr mortality for hospital survivors is described. Administrative data from April 1, 1994–March 31, 1999 was used to study the entire scope of rural and urban hospital practice in one Canadian province.
Methods
Defining pneumonia and data extraction
Health-service databases used included the following: 1) Canadian Institute for Health Information's (CIHI) Inpatient Discharge Abstract Database (DAD) for the province of Alberta for 1994–1995 to 1998–1999; 2) Alberta Physician Claims Assessment System Database for 1994–1995 to 1999–2000; and 3) Alberta Health Insurance Plan Registry File for 1994–2000. The analysis was performed within the protected environment of Alberta Health and Wellness, which is governed by provincial legislative guidelines on the confidentiality of health information. These data capture nearly the entire population and included a unique, anonymous personal identifier allowing linkage between databases.
Inclusion criteria were CAP, defined as the most responsible diagnosis (MRD) or any of the other 15 diagnosis codes defined to be type 1 (the diagnosis existed prior to admission) with the International Classification of Disease, ninth revision (ICD‐9), Clinical Modification (CM) values of 480–487.x (pneumonia) or 507.x (aspiration pneumonia), excluding 484.x (pneumonia of infectious diseases classified elsewhere) found in CIHI DAD 13, 14.
Exclusion criteria included: 1) not an Alberta resident or not treated in an Alberta acute-care facility; 2) an adjacent diagnosis with hospitalisation for a surgical procedure; and 3) any previous hospitalisation <10 days since incident pneumonia case.
Physician visits for CAP were extracted from the Alberta Physician Claims Assessment System Database. Physician visits were defined as consultation claims (Canadian Classification of Procedures (CPX) 03.x) for any diagnosis 2–14 days prior to admission date 15. The MRD was classified as CAP (as defined above), any respiratory diagnosis (ICD‐9‐CM 460–519.9, excluding pneumonia), or any other diagnosis. All physician characteristics were extracted from hospital abstracts.
Patient factors
Recorded demographical and comorbidity factors included the following: 1) age (18–44, 45–64, 65–74, 75–84, or ≥85 yrs (age at the fiscal year end is used )); 2) sex; 3) aboriginals registered with the Dept of Indian Affairs and Northern Development, recorded within the Alberta Healthcare Premium Registry databank; 4) urban or nonurban region of residence; and 5) Charlson's comorbidities, extracted using Deyo's criteria and each of the 12 diagnostic categories compiled as none, one, two and more than two 16, 17.
Recorded severity-of-illness factors included the following: 1) transfer to hospital from a nursing home, long-term care or continuing-care institution; 2) transfer to hospital by ambulance as defined in hospital abstract; 3) admission from emergency room as defined in hospital abstract; 4) emergent admission from home/office/clinic as defined in hospital record abstract; 5) special (intensive) care unit admission (as defined by each hospital); 6) diagnosis code of respiratory failure or arrest (ICD‐9‐CM 518.81, 799.1); 7) diagnosis code of hypotension or shock (ICD‐9‐CM 458.xx, 758.5x); 8) procedure code for ventilation >96 h (ICD‐9‐CM 96.72); and 9) procedure code for dialysis (ICD‐9‐CM 39.95, 54.98). Severity of presentation was defined as any one of special-care unit admission, hypotension/shock, in-hospital dialysis, respiratory arrest/failure or ventilation for >96 h.
Physician factors included the following: 1) general/family practitioner listed as most responsible hospital physician; 2) any internal medicine specialist or subspecialist listed as consultation physician; 3) annual volume of pneumonia cases per physician (by quartile rank), including any hospital consultations and most responsible hospital physician; 4) pre-admission ambulatory visit (physician claims CPX 03.x) by any physician 2–14 days prior to admission date with the diagnosis code for pneumonia as defined above; 5) pre-admission ambulatory visit (physician claims CPX 03.x) by an internal medicine specialist 2–14 days prior to admission date with the diagnosis code for pneumonia as defined above; and 6) any emergency room visit 2–14 days prior to admission date with the diagnosis code for pneumonia as defined above. Physician specialty was coded in the hospital abstract.
Healthcare organisation factors included the following: 1) admission to a hospital not within the patient's defined health region of residence (see below); 2) transfer from a smaller to a larger acute-care facility (see below) for the diagnosis of pneumonia as defined above; 3) remote distance to hospital (see below); 4) hospital volume (see below); 5) hospital beds per capita in the hospital's health region (all active acute-care hospital beds in each year per health region were surveyed and maintained in the provincial databases); 6) bed turnover defined as the ratio of all hospital admissions per hospital bed in the specific hospital on the date of pneumonia admission; 7) occupancy rate defined as the daily census per hospital bed in the specific hospital at the date of pneumonia admission; 8) weekend (Saturday/Sunday/statutory holiday) admission date; 9) admission time 18:00–08:00 h; and 10) admission in December 1‐March 31 (peak season for pneumonia hospital admissions in Alberta).
Derived variables
Defining hospital type by patient volume
All hospitals in Alberta are administered by an autonomous regional board in each of the 17 health regions. All 3 million residents of Alberta are enrolled in the public healthcare insurance plan and reside in one of the 17 health regions. Hospitals admitting patients with CAP were categorised into five groups on the basis of the average number of hospital discharges per year over the 5‐yr study period, geographic location and medical-school proximity.
Rural hospitals included 77 hospitals with <50 cases·yr−1 and 27 hospitals with 50–108 cases·yr−1. The rural hospital cut-off values for pneumonia cases of 50 and 108 represented the 50th and 75th percentiles. Regional hospitals included five hospitals located in one of the five nonmetropolitan, regional, healthcare cities with 67–251 cases·yr−1, and one high-volume rural hospital with 221 cases·yr−1. Metropolitan hospitals included seven hospitals located in the metropolitan health regions of Calgary and Capital hospitals with 92–813 cases·yr−1 and two hospitals located adjacent to medical schools; one hospital in each of two metropolitan centres with 493 and 610 cases·yr−1.
Calculating remote distance to hospital
Each separation was mapped to the centre of a postal code and the distance “as the crow flies” between centroids calculated. Nearest hospital distance to resident postal code was obtained for all nonurban residents. Urban resident distances were zero. Distances >50 km were defined as remote.
Mortality
Vital-statistics declaration of death certificates were merged with the Alberta Healthcare Premium Registry databank in order to determine deregistration from the plan. Reliability of linkage was high (99.5%), verified by comparing in-hospital deaths, recorded by hospital-chart extraction, to Alberta Healthcare Premium Registry death, derived from vital statistics.
Outcomes used in regressions included the following: 1) in-hospital death per hospital discharge within the first 72 h of admission; 2) in-hospital mortality per hospital discharge; and 3) 1‐yr mortality per hospital discharge in those not dying in hospital.
Statistics
A forward, stepwise selection of factors was used to determine the order of importance of the factor influencing mortality. The factors were grouped as patient severity of illness and comorbidity group, healthcare organisation group or physician group. Each entire factor group was also sequentially forced in the model and the c‐statistic determined. The c‐statistic is a summary-rank correlation statistic between the observed responses and predicted probabilities that equals the area under the receiver-operator curve and is commonly reported for risk-adjustment models 18.
The unit of analysis was hospital discharge, with some patients being hospitalised on multiple occasions per year. The authors attempted to adjust for correlations in hospital discharges within the same subject, but it did not result in a substantial difference from those under the assumption of zero correlation. Therefore, the authors opted to report results of a simpler approach based on multiple logistic models 19, 20. To control the large-sample size effect on statistical significance, a one-third random sample from the hospital discharge data was used for statistical modelling 21. The results were validated with the entire data set to test for bias in the sample. Significance was defined as p<0.05.
Results
During the 5 yrs of the study there were 43,642 acute-care pneumonia hospital discharges. The in-hospital mortality per hospital discharge was 11% (9.2% in the first 24 days as shown in figure 1⇓) and the overall 1‐yr mortality per hospital discharge was 26%. Of 4,693 in-hospital deaths, 11.5% (538) died within 24 h of admission, 16.5% (773) within 25–72 h of admission, 45.0% (2,111) within 73 h to 2 weeks of admission and 27.1% (1,271) died >2 weeks after admission.
In-hospital survival for community-acquired pneumonia.
Mortality (table 1⇓) was proportionally higher in males, older patients, those of urban residence and those with comorbidity (in particular malignancy, congestive heart failure, vascular diseases, dementia and renal disease; these data not shown). The severity-of-illness markers identified proportionally higher mortality, except for seasonal admissions (table 2⇓). Severity of illness tended to be greater for deaths within 72 h compared with all in-hospital deaths. Mortality was proportionally greater in urban hospitals and hospitals with proportionally greater mortality were located closer to patients' home residences (table 3⇓). The median daily hospital occupancy rate was 75% (interquartile range 56–90%). The median number of daily admission per bed was 0.15 (0.1–0.2). Mortality was proportionally greater when internal medicine specialists/subspecialists were involved in care, either as the most responsible physician or consultant (table 4⇓). Only 5% of patients had a most responsible physician that was neither a general/family practitioner nor an internal medicine specialist/subspecialist.
Patient demographical and comorbidity factors for hospitalised community-acquired pneumonia in Alberta during April 1, 1994–March 31, 1999
Severity of illness for hospitalised community-acquired pneumonia in Alberta during April 1, 1994–March 31, 1999
Hospital factors for community-acquired pneumonia in Alberta during April 1, 1994–March 31, 1999
Physician factors for hospitalised community-acquired pneumonia in Alberta during April 1, 1994–March 31, 1999
The mean number of hospitalised CAP patients per most responsible unique physician was 7.4 patients·yr−1. These were grouped into quartiles (first quartile: 0–6 patients·yr−1; second quartile: 7–13 patients·yr−1; third quartile: 14–27 patients·yr−1; and fourth quartile: >27 patients·yr−1). Mortality was proportionally greater in those without a physician consultation 2–14 days prior to admission. Physician consultations prior to hospitalisation were for the diagnosis of pneumonia (7%), all respiratory diseases (17%) and all other diagnosis (76%). Mortality was proportionally greater in those admitted directly from the emergency room, but not in those with an emergency visit 2–14 days prior to admission.
Internal medicine specialists or subspecialists and high-volume physicians cared for patients with greater severity of presentation and more comorbidity (table 5⇓). Similarly, pre-admission consults by internal medicine specialists or subspecialists (but not for general practice physicians) occurred in patients with greater severity of presentation and more comorbidity. General practice physicians constituted 84, 88, 80 and 19%, respectively, of each of the four physician volume quartiles (from low-to-high volume, respectively).
Relationship between severity of illness, comorbidity and physicians for community-acquired pneumonia in Alberta during April 1, 1994–March 31, 1999
The c‐statistics for pneumonia mortality models ranged 0.77–0.79, where a value of 1 describes perfect prediction and 0.5 describes prediction no better than by chance (⇓⇓tables 6–8⇓) 18. The patient factor group provided the best prediction of the mortality probabilities within 72 h, in terms of model entry order and c‐statistic. Physician factors associated with decreased mortality within 72 h were a nonspecialist, most responsible physician and specialist consult (during or prior to admission). None of the hospital-care organisation factors were significant.
Logistic analysis for hospital death within the first 72 h of admission
Logistic analysis for in-hospital death
Logistic analysis for death at 1 yr excluding all in-hospital deaths
Similar results were noted for all in-hospital deaths, with respect to patient factors (table 7⇑). In this case, physician factors demonstrated decreased in-hospital mortality, with higher patient volumes for the most responsible physicians. Consults with an internal medicine specialist/subspecialist prior to admission were still associated with decreased in-hospital mortality, but in-hospital consultation was associated with increased in-hospital mortality. Compared with the patient factor group, the physician factor group provided less power of model prediction in terms of model-entry order and c‐statistic. Healthcare organisation factors (timing of admission) became significant in predicting in-hospital death.
Patient factors provided the greatest power of prediction for 1‐yr mortality of those discharged from hospital (table 8⇑). In-hospital specialist/subspecialist and high-volume physicians were associated with increased mortality. Healthcare organisation factors were significant but had the least predicting power of death within a year. Hospital admission, not region of residence, was associated with increased mortality, while admissions to larger volume or metropolitan hospitals were associated with decreased mortality.
Discussion
This study explored the association of mortality with patient factors, physician factors and healthcare organisation factors after admission to hospital with CAP. Mortality rates of 11% occurred in hospital and 26% at 1 yr 3. As expected, severity of illness and comorbidity had the strongest association with mortality 22. This paper is the first study associating high-volume physician care with decreased in-hospital mortality for CAP. The presence of an internal medicine specialist/subspecialist was not consistently associated with decreased hospital mortality except when a pre-admission consultation occurred. Healthcare organisation factors were weakly associated with in-hospital mortality for the acute episode and mortality in those that were discharged after the initial index episode of pneumonia. One-year mortality was decreased in high-volume hospitals, where the original admission occurred.
The relationships between specialists/subspecialists, high-volume practice and more complex (greater severity of illness and comorbidity) patients are not easily separable. Specialists/subspecialists practice in higher volume centres and see the more complex patients on referrals. Thus, any remaining confounding, due to patient selection bias towards severe illness/comorbidity, would tend to lower mortality associated with general practice physicians, while it increases mortality associated with specialists/subspecialists and high-volume physicians.
Randomised trials comparing high/low volume and general/specialist practice are logistically difficult considering the need for large sample sizes to detect the small effects of practice and need to locate geographical areas where such a mix of physicians exist. Administrative-database studies capture the full spectrum of practices and patients, with sufficient sample size to demonstrate an effect. This study does not support the contention that in-hospital specialist/subspecialist care resulted in a unique mortality advantage. However, it is supportive of an advantage for high-volume practice. Many studies addressing the relationship between volume of procedures and patient outcomes have been published, most with respect to surgical care 23–25. Similar to the current study, hospital size or the physician volume was used as a proxy for physician skill. CAP, unlike surgical volume studies, is not associated with a unique procedure or skill. Treatment of pneumocystis pneumonia in human immunodeficiency virus-seropositive patients has been associated with improved outcomes when treated in higher volume centres 26. Among studies analysing hospital volumes and outcomes, better outcomes have been associated with higher patient volumes in some instances 27–30, but not others 28, 31, 32. Some studies of individual-provider volume have shown a positive relationship between volume and outcomes 33, 34, whereas others have shown no relation or inconsistent results 23, 35. Finally, only a few studies similar to the current study have analysed both hospital volume and provider volume and reported a positive volume/outcome relation 36, 37.
This study does not specify which aspects of high-volume care may be associated with decreased mortality. Specialist/subspecialist evaluation and treatment may be better because of greater conformity to published guidelines 38, 39. It is not known if high-volume practice conforms to treatment guidelines for CAP. Early physician care documented as a pre-admission visit was associated with decreased in-hospital mortality, suggesting that expedient medical care is an important component of a care pathway for CAP 2. Physicians were specified as the clinician involved with high-volume practice. Other healthcare workers may also become more skilled where high-volume practice is supported. However, hospital volume itself was not associated with lower mortality. The current study, unlike others 40, did not find any association with decreased mortality, at a time when hospital staffing, including consultants, allied healthcare specialties and management, may be more limited (weekends and holidays, 18:00–08:00 h). Another possible advantage to high-volume care or pre-admission consultation is timely, appropriate, antibiotic administration 41, 42. Finally, high-volume practitioners may simply be better able to anticipate problems, rather than react postoccurrence. The ability to emulate high-volume practice may represent another strategy to help physicians follow practice guidelines 43. This may be particularly important since most cases of hospitalised pneumonia are cared for by primary-care physicians in community hospitals 44. This study does not specify why some of the hospital organisation factors were associated with decreased mortality for CAP mortality. The benefit of high-volume hospitals may be related to better aftercare in the old and frail, which are at greater risk of death, rather than the specific treatment of the actual pneumonia.
The current study has several limitations. Pneumonia may exacerbate a pre-existing comorbidity. The authors did not know which came first (i.e. the pneumonia made the comorbidity worse or the comorbidity became worse and pneumonia ensued). As the patient cohort was older, almost all patients had a least one comorbidity. The authors do not believe the data was sufficiently detailed to accurately and reliably identify all internal medicine subspecialties as distinct from internal medicine speciality. This is especially true as some physicians have multiple-board certification and the physician reimbursement rules sometimes result in the same individual physician using different specialty designation. Data from the current study was sufficiently robust to identify specialists/subspecialists from nonspecialists, since this indicator has important financial implications for both the practitioner and hospital.
Population-based administrative database research may be highly generalised, although limited in clinical details. The authors attempted to adjust for case severity and comorbidity, but may not have captured all variation 45. These variables are likely to be less reliable than a clinically derived pneumonia index 2. Processes of care variables, such as use of a special/intensive care unit or transfer to another hospital, are not justified by a specific threshold. As such, these variables would be expected to vary widely between patients and hospitals. Despite these limitations, the model's overall prediction of mortality was good as demonstrated by the c‐statistic. The authors were not able to uniquely identify physicians who practiced in more than one hospital. Therefore, combined case volumes could not be credited to these multi-institutional physicians.
- Received December 12, 2002.
- Accepted February 28, 2003.
- © ERS Journals Ltd
References
