This Article Abstract Full Text (PDF) All Versions of this Article: 200706-813OCv1 177/9/989    most recent Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Lin, H.-C. Articles by Hwang, Y.-T. Search for Related Content PubMed PubMed Citation Articles by Lin, H.-C. Articles by Hwang, Y.-T.

Physician's Case Volume of Intensive Care Unit Pneumonia Admissions and In-Hospital Mortality

¹ School of Health Care Administration, Taipei Medical University, Taipei, Taiwan; ² Arnold School of Public Health, Department of Health Services Policy and Management, University of South Carolina, Columbia, South Carolina; ³ Intensive Care Unit, Taipei Medical University Hospital, Taipei, Taiwan; and ⁴ Department of Statistics, National Taipei University, Taipei, Taiwan

Correspondence and requests for reprints should be addressed to Herng-Ching Lin, Ph.D., School of Health Care Administration, Taipei Medical University, 250 Wu-Hsing Street, Taipei 110, Taiwan. E-mail: henry11111{at}tmu.edu.tw

	ABSTRACT

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 
Rationale: Although several studies have investigated volume–outcome relationships for surgical procedures, there has been no such study of intensive care unit (ICU) patients admitted for pneumonia.

Objectives: This study examines associations between in-hospital mortality of ICU-admitted pneumonia patients and their attending physician's case volume.

Methods: We used 2002–2004 claims data from Taiwan's National Health Insurance for all 87,479 adult ICU admissions for pneumonia. Patients were assigned to one of four groups, on the basis of their physician's ICU pneumonia case volume (low volume, <36 cases; medium volume, 37–114 cases; high volume, 118–314 cases; and very high volume, 315 cases). Generalized estimating equations (conditional on hospital, and unconditional) were used, adjusting for physician demographics and specialty, hospital characteristics, patient characteristics (including clinical severity and comorbidities), and physician-level random effect (clustering effect) to assess whether physicians' case volume predicts in-hospital mortality.

Measurements and Main Results: In-hospital mortality systematically declined with increasing physician case volume: 14.7, 14.3, 11.4, and 8.1% from low-volume to very-high-volume groups. Adjusted unconditional odds of mortality among low-volume physicians' patients were 2.04 times those of very-high-volume physicians, 1.35 times that of high-volume physicians, and 1.09 times those of medium-volume physicians (all P < 0.001). The relationship is sustained when the odds are estimated conditional on hospital, when initial 5-day mortality is separated from 30-day mortality, and when pulmonologists' and critical care specialists' patients are studied separately.

Conclusions: Physician volume significantly predicts inpatient mortality among ICU patients with pneumonia. Detailed study of clinical approaches, decision algorithms, and treatment plans of high-volume physicians is recommended to identify possible mediating factors in this phenomenon.

Key Words: intensive care unit • pneumonia • outcome assessment

AT A GLANCE COMMENTARY TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES   Scientific Knowledge on the Subject Although several studies have investigated volume–outcome relationships for complex surgical procedures, there has been no outcomes study of intensive care unit patients admitted for pneumonia. What This Study Adds to the Field This study found that physician volume significantly predicts inpatient mortality among intensive care unit patients with pneumonia.

 
Worldwide, pneumonia is the most frequent cause of communicable disease admissions and inpatient deaths. An estimated 5.6 million cases of community-acquired pneumonia occur annually in the United States (1), with 1.2 million patients hospitalized, and with an inpatient mortality rate of 5.8% (2).

Attending physicians' case volume as a predictor of care outcomes is widely documented, particularly for complex medical conditions and surgical procedures, such as cancer care, coronary artery bypass surgery, and others (3–5). Little documentation on pneumonia case volumes and outcomes is available. Such studies could enable provider-driven and payer-driven strategies to reduce unwarranted variations in outcomes.

Durairaj and colleagues reported that medical intensive care units (ICUs) with high respiratory patient volumes had lower risk-adjusted mortality for sicker patients, based on a 29-hospital study in Ohio (6). Respiratory diagnoses could span a heterogeneous range of conditions of diverse origins (e.g., infectious, malignant, degenerative). Therefore, there could be considerable confounding, yielding limited policy and practice implications. Our study uses inpatient claims data from Taiwan with a primary diagnosis of pneumonia and an ICU stay. In Taiwan, all teaching (and many nonteaching) hospitals have state-of-the-art clinical practice guidelines and standard orders set in place for pneumonia (7). We used nationwide, 3-year, population-based (de-identified) data from Taiwan's National Health Insurance (NHI) claims database, released for research purposes, to examined whether physicians with higher case volumes offer ICU patients with pneumonia better outcomes.

	METHODS

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 
Database
We used inpatient claims data for 2002–2004, which include data on every inpatient admission among NHI beneficiaries, more than 21 million citizens (96% of Taiwan's population). The database has ICD-9CM (International Classification of Diseases, Ninth Revision, Clinical Modification) codes for the principal diagnosis, and up to four secondary diagnoses. The NHI Bureau implements routine sample crosschecks of each hospital's claims with medical charts, followed by punitive measures for coding infractions, which deters diagnosis up-coding. This deterrent is counterbalanced by the NHI's reimbursement system, which ties a hospital's reimbursement level to its patient severity profile. As a result, hospitals' interests are best served by accurate coding of diagnoses and care items. Although there are no documented sensitivity and specificity studies of coding accuracy, it is generally believed that the NHI's checks and balances foster accurate coding.

Taiwan's NHI provides universal coverage to all citizens, a single plan with generous benefits, low copayments, and free choice of a widely dispersed network of public, private, and not-for-profit providers. These features make these population-based data a valuable resource to examine risk-adjusted outcome variations across a variety of provider characteristics, with minimal confounding.

Study Sample
We selected adult patients admitted with a principal diagnosis of pneumonia (ICD-9-CM: 480–483.8, 485–487.0), with an ICU stay during hospitalization. Of more than 6 million inpatient claims during January 2002 to December 2004, 479,715 were pneumonia admissions; 101,105 of these patients were admitted to ICUs. Of these, 7,384 pediatric cases (<18 yr) and 5,971 cases transferred to or from another hospital were excluded, for a final study sample of 87,479 cases.

Classification of Patients by Their Physician's Case Volume
Using attending physician identifiers in the claims, we classified patients by their physician's total case volume of patients with pneumonia with an ICU stay during the study period. Physicians were sorted, in ascending order of volume, and volume cutoff points were determined, such that the sample was classified into four, approximately equal-sized groups (consistent with the documented methodology for such studies [8–10]). The four volume groups were as follows: low volume (36 cases treated by their attending physician), medium volume (37–117 cases), high volume (118–314 cases), and very high volume (315 cases). The respective volume groups had 21,927, 21,928, 21,883, and 21,741 cases.

Statistical Analysis
The SAS package (version 9.0; SAS Institute, Cary, NC) was used. The outcome measure was "in-hospital mortality," with "patient" as the unit of analysis and "physician volume" measured at physician level. A generalized estimating equation (GEE; unconditional) was used to examine the adjusted relationship of physician volume with in-hospital mortality, controlling for physician, hospital, and patient characteristics, and accounting for clustering of patients within physician panels (physician-level random effect). We also performed a conditional (fixed effects) logistic regression model in which observations are conditioned on hospital, to partition out systematic hospital-specific variation, to address potential confounding by a possible association between high-volume physicians and certain types of hospitals (e.g., teaching hospitals) (11). Finally, a sensitivity analysis was performed to test for the consistency of the volume–outcome relationship under various volume cutoff points.

We adjusted for the following: physician's sex, age, and specialty (pulmonology or critical care medicine vs. another specialty); the hospital's ICU bed strength, its teaching status, and ownership (public, for-profit, and not-for-profit); patient demographics (age and sex); and patient severity. Patient severity was measured by clinical severity of the pneumonia and presence of comorbidities. The proxy for pneumonia severity was twofold: duration of mechanical ventilation use when it was used, as shown in Table 1, and presence of acute respiratory failure (ARF) without significant duration of mechanical ventilation (3,111 patients either rapidly succumbed to the respiratory failure or responded to other medical treatment). Clinical comorbidity was captured using the Charlson Comorbidity Index (CCI) (12), which is used to capture mortality risk in administrative claims data analyses. The CCI is the sum of the patient's weighted scores of the relative mortality risk of 19 conditions (e.g., congestive heart failure, myocardial infarction, liver disease, cancer, dementia) with a score of zero if no comorbid conditions exist.

	RESULTS

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 
The study sample distribution by patient variables is shown in Table 1: 63.5% were male, 62.8% were 74 years or older, 65.7% had pneumonia complicated with ARF, and 62.1% received mechanical ventilation during hospitalization. Mean length of stay and hospitalization costs were 24.8 days (± 18.1 d) and Taiwan New Dollars (TWD $) 281,508 (±TWD $228,772; US $1 = TWD $33 in 2004), respectively.

The bivariate sample distribution by physician volume and physician characteristics is shown in Table 2. For the total 3,667 attending physicians of the sample patients, the mean case volume was 25.8 patients. Very-high-volume physicians were more likely to be male and specialize in pulmonary or critical care medicine (both P < 0.001), and have older patients, with lower average CCI scores and higher incidence of ARF and mechanical ventilation use, compared with physicians in other volume groups (all P < 0.001).

Table 4 shows the association of physician volume with in-hospital mortality among the subsample (50,064 patients) treated by pulmonologists and critical care specialists. The relationship is sustained: low-volume specialists' patients have about twice the risk-adjusted mortality odds as those of very-high-volume and high-volume physicians (OR, 1.89 and 2.13, respectively).

	DISCUSSION

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 
This article reports on the volume–mortality relationship for adult ICU pneumonia cases on the basis of nationwide, population-based data, covering 87,479 patients treated by 3,667 physicians. Adjusted for relevant confounders, and robust to a sensitivity analysis, low-volume physicians' patients have twice the mortality odds of patients of very-high-volume physicians. Importantly, the finding holds even among the subsample of patients treated by pulmonologists and critical care specialists. (However, within the total sample, these specialists have clearly lower adjusted mortality odds [0.66] relative to other specialties.)

Conditional logistic regression analysis eliminates potential confounding by unmeasured hospital variables on many counts. First, high-volume physicians likely practice at better equipped and/or teaching hospitals, which could potentially account for the superior outcomes of high-volume physicians, due to these hospitals' infrastructure and staffing advantages. Conditional logistic regression essentially evaluates the volume–outcome association among physicians within each hospital and then averages these effects across hospitals. We found that the physician volume–outcome association is sustained in the conditional logistic regression model, although some mitigation of the ORs is observed. The latter suggests that intrahospital variables do mediate part of the physician volume–outcome relationship, but the magnitude of the drop also shows that most of the mortality differentials are driven by the independent effect of the physicians' case volume (experience), regardless of the hospital in which they practice. Our findings suggest that previously documented associations between hospital volume and mortality (13–15) may have been confounded by more experienced physicians practicing in larger volume hospitals.

Durairaj and coworkers reported lower risk-adjusted mortality among ICU patients with a respiratory diagnosis who were admitted to high-volume hospitals compared with those admitted to low-volume hospitals (6), and many studies show that higher ICU patient volumes are associated with superior outcomes among critically ill patients (13–18), all consistent with our findings.

Our study contributes to the literature by demonstrating that, after isolating the effect of most institutional and patient confounders, the attending physician's volume independently predicts mortality in severely ill patients with pneumonia. The adequacy of our adjustments for patient severity could be questioned. Several study attributes mitigate this concern. First, we restricted our study sample to inpatients with an admission diagnosis of pneumonia and who needed ICU care, itself an indictor of significantly higher severity. Second, we adjusted for ARF, which significantly drives mortality in pneumonia cases. We used specific operational variables that mediate the role of ARF in mortality outcomes: whether the patient survived the initial onset of ARF, and the duration for which mechanical ventilation had to be used. We have accounted for both these variables, captured as ARF without mechanical ventilation, and duration of continuous use of mechanical ventilation among survivors. These two variables, together with our sample selection criterion of an ICU stay should capture most of the significant impact of pneumonia severity, and may have mitigated the utility of the Pneumonia Severity Scale in this study.

In our sample, patients of high-volume physicians had higher incidence of ARF, and received longer durations of mechanical ventilation, indicating higher critical severity. Therefore, differential case severity should have aggravated high-volume physicians' outcomes, not improved them. Our adjustment for these variables emphasizes the superior outcomes of high-volume physicians. Another source of confounding could be unmeasured differences in clinical severity between the volume groups, such as pre-pneumonia conditions, hemodynamic status, hyperglycemia, and others. It is highly implausible that these factors were systematically higher among low-volume physicians' patients. Moreover, because patients with pneumonia are admitted to an ICU when their condition is clinically severe enough, it is unlikely that systematic unmeasured differences in severity exist between low-volume and high-volume physicians' patients in our sample.

Three explanations could be hypothesized for the volume–outcome relationship (each having different policy implications):

1. High-volume hospitals may implement systems-based practices that improve outcomes (e.g., multidisciplinary care rounds, high nurse-to-patient ratios).
   
2. Practice makes perfect (i.e., physicians who accumulate experience with caring for certain types of patients provide and direct better care, causing improved outcomes).
   
3. Selective patient self-referral to reputed hospitals or physicians may cause better outcomes through two mechanisms: higher volumes of these reputed hospitals confound their inherently superior outcomes as a volume effect, projected to its physicians essentially a halo effect. Patient self-referral also implies that those with less severe disease can "shop for" a "reputed" hospital regardless of distance, whereas the severely ill settle for the nearest hospital (in Taiwan, patients have full choice of any physician or hospital in the country, without referral). Such selection bias should cause high-volume hospitals to have disproportionately less severe patients and therefore better outcomes.
   

If hypothesis 1 is empirically sustained, then smaller hospitals should be required to adopt the systems-based practices of larger hospitals. If hypothesis 2 is supported, payers should consider reimbursing consultations with an experienced physician until each physician completes a threshold number of ICU pneumonia cases, regardless of specialty status. Our study confirms that, regardless of the hospital in which they practice, high-volume physicians perform better. Disproportionately higher severity among high-volume physicians' panels (age, ARF, and duration of mechanical ventilation) argues against selection bias due to selective self-referral.

Overall, our results suggest that practice makes perfect, supporting the latter policy implication. Cost of care differences alone may justify such expense, apart from being potentially life-saving. At the same time, a statistically significant contribution by the hospital effect (reduction in mortality odds in the conditional model) implies that payers should consider requiring all hospitals to implement the systems-based practices of larger hospitals, as well as regionalization of care for severe pneumonia cases.

Finally, the consistency of the volume–outcome relationship among the subsample of respiratory specialists' patients and among the subsample with ARF, and on separating early 5-day mortality from delayed, 6–30-day mortality, all suggest strong support for the "practice makes perfect" hypothesis. Other studies concerning volume–outcome relationships for other diagnoses/procedures in Taiwan also consistently demonstrate that physician volume is a more critical factor for patient outcome than hospital volume (19–21). It has been suggested that high-volume physicians can better manage unexpected problems and medical situations (7).

This study has several strengths. Because it used a nationwide population-based dataset from a single-payer insurance system, its findings are noteworthy for clinicians and policy makers, who often have to contend with indeterminate findings from studies using data from pluralistic health systems with multiple insurers, and limited patient panels. Second, its large sample size has provided ample statistical power to detect differences between study groups, after adjusting for many confounding variables.

There are some study limitations. The accuracy of comorbidity documentation is generally questioned when claims data are used for research. As indicated earlier, there are built-in checks and balances in the NHI's oversight and reimbursement system. Measures such as routine sampling of patient charts to crosscheck with claims, detailed audits for outlier hospitals, and punitive fines for fraudulent claims deter diagnoses and care up-coding. Concurrently, the reimbursement system favors accuracy in the opposite direction. Because each hospital's reimbursement rate is tied to its patient severity profile, the incentive is to ensure accurate and complete coding of comorbidities to realize a severity-appropriate reimbursement rate. Therefore, it is generally believed that coding is largely accurate, although the lack of systematic sensitivity analyses documentation makes this issue a potential study limitation.

	Conclusions

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 
There are several policy implications. First, similar studies need to be replicated in other countries and settings, examining diagnosis-specific volume–outcome relationships in ICUs for pneumonia. Second, policy makers should consider routinely paying for second-opinion consultations with experienced physicians until each physician crosses a critical threshold of case volume. This should save considerable costs as well as lives. Third, our supplementary analysis showed higher costs incurred toward aggressive treatment procedures among patients of high-volume physicians, and higher drug and lab diagnostic costs (results not presented). This finding suggests that much will be gained from clinician-led peer reviews of the differences in clinical approaches and care processes between high-volume and low-volume physicians. This should enable more detailed explication of decision algorithms for use by all providers.

	APPENDIX 1.

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 

	APPENDIX 2.

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

	FOOTNOTES

Conflict of Interest Statement: None of the authors has a financial relationship with a commercial entity that has an interest in the subject of this manuscript.

Received in original form June 4, 2007; accepted in final form January 31, 2008

	REFERENCES

TOP ABSTRACT AT A GLANCE COMMENTARY METHODS RESULTS DISCUSSION Conclusions APPENDIX 1. APPENDIX 2. REFERENCES

 

1. Bartlett JG, Dowell SF, Mandell LA, File TM Jr, Musher DM, Fine MJ. Practice guidelines for the management of community-acquired pneumonia in adults. Infectious Diseases Society of America. Clin Infect Dis 2000;31:347–382.[CrossRef][Medline]
2. Lindenauer PK, Behal R, Murray CK, Nsa W, Houck PM, Bratzler DW. Volume, quality of care, and outcome in pneumonia. Ann Intern Med 2006;144:262–269.[Abstract/Free Full Text]
3. Luft HS, Bunker JP, Enthoven AC. Should operations be regionalized? The empirical relation between surgical volume and mortality. N Engl J Med 1979;301:1364–1369.[Abstract]
4. Flood AB, Scott WR, Ewy W. Does practice make perfect? Part I: the relation between hospital volume and outcomes for selected diagnostic categories. Med Care 1984;22:98–114.[Medline]
5. McGrath PD, Wennberg DE, Malenka DJ, Kellett MA Jr, Ryan TJ Jr, O'Meara JR, Bradley WA, Hearne MJ, Hettleman B, Robb JF. Operator volume and outcomes in 12,998 percutaneous coronary interventions: Northern New England Cardiovascular Disease Study Group. J Am Coll Cardiol 1998;31:570–576.[Abstract/Free Full Text]
6. Durairaj L, Torner JC, Chrischilles EA, Vaughan Sarrazin MS, Yankey J, Rosenthal GE. Hospital volume-outcome relationships among medical admissions to ICUs. Chest 2005;128:1682–1689.[CrossRef][Medline]
7. Hwang KL, Su WW, Soon MS, Wu SS, Chen YY, Chen MY. Evaluation of the implementation of clinical guidelines for transcatheter arterial chemo-embolization of patients with hepatocellular carcinoma [in Chinese]. Taiwan J Public Health 2005;24:284–295.
8. Hannan EL, Racz M, Ryan TJ, McCallister BD, Johnson LW, Arani DT, Guerci AD, Sosa J, Topol EJ. Coronary angioplasty volume-outcome relationships for hospitals and cardiologists. JAMA 1997;277:892–898.[Abstract/Free Full Text]
9. Jollis JG, Peterson ED, Nelson CL, Stafford JA, DeLong ER, Muhlbaier LH, Mark DB. Relationship between physician and hospital coronary angioplasty volume and outcome in elderly patients. Circulation 1997;95:2485–2491.[Abstract/Free Full Text]
10. Birkmeyer JD, Siewers AE, Finlayson EV, Stukel TA, Lucas FL, Batista I, Welch HG, Wennberg DE. Hospital volume and surgical mortality in the United States. N Engl J Med 2002;346:1128–1137.[Abstract/Free Full Text]
11. Localio AR, Berlin JA, Ten Have TR, Kimmel SE. Adjustments for center in multicenter studies: an overview. Ann Intern Med 2001;135:112–123.[Abstract/Free Full Text]
12. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373–383.[CrossRef][Medline]
13. Glance LG, Li Y, Osler TM, Dick A, Mukamel DB. Impact of patient volume on the mortality rate of adult intensive care unit patients. Crit Care Med 2006;34:1925–1934.[CrossRef][Medline]
14. Kahn JM, Goss CH, Heagerty PJ, Kramer AA, O'Brien CR, Rubenfeld GD. Hospital volume and the outcomes of mechanical ventilation. N Engl J Med 2006;355:41–50.[Abstract/Free Full Text]
15. Phibbs CS, Baker LC, Caughey AB, Danielsen B, Schmitt SK, Phibbs RH. Level and volume of neonatal intensive care and mortality in very-low-birth-weight infants. N Engl J Med 2007;356:2165–2175.[Abstract/Free Full Text]
16. Marcin JP, Song J, Leigh JP. The impact of pediatric intensive care unit volume on mortality: a hierarchical instrumental variable analysis. Pediatr Crit Care Med 2005;6:136–141.[CrossRef][Medline]
17. Jones J, Rowan K. Is there a relationship between the volume of work carried out in intensive care and its outcome? Int J Technol Assess Health Care 1995;11:762–769.[Medline]
18. Iapichino G, Gattinoni L, Radrizzani D, Simini B, Bertolini G, Ferla L, Mistraletti G, Porta F, Miranda DR. Volume of activity and occupancy rate in intensive care units: association with mortality. Intensive Care Med 2004;30:290–297.[CrossRef][Medline]
19. Wen HC, Tang CH, Lin HC, Tsai CS, Chen CS, Li CY. Association between surgeon and hospital volume in coronary artery bypass graft surgery outcomes: a population-based study. Ann Thorac Surg 2006;81:835–842.[Abstract/Free Full Text]
20. Lien YC, Huang MT, Lin HC. Association between surgeon and hospital volume and in-hospital fatalities after lung cancer resections: the experience of an Asian country. Ann Thorac Surg 2007;83:1837–1843.[Abstract/Free Full Text]
21. Xirasagar S, Lien YC, Lin HC, Liu TC, Tsai J. Procedure volume of gastric cancer resections versus 5-year survival. Eur J Surg Oncol 2008;34:23–29.[Medline]

This article has been cited by other articles:

				R. A. Fowler, N. K. J. Adhikari, D. C. Scales, W. L. Lee, and G. D. Rubenfeld Update in Critical Care 2008 Am. J. Respir. Crit. Care Med., May 1, 2009; 179(9): 743 - 758. [Full Text] [PDF]

This Article Abstract Full Text (PDF) All Versions of this Article: 200706-813OCv1 177/9/989    most recent Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Lin, H.-C. Articles by Hwang, Y.-T. Search for Related Content PubMed PubMed Citation Articles by Lin, H.-C. Articles by Hwang, Y.-T.