Published ahead of print on September 14, 2006, doi:10.1164/rccm.200511-1810OC
© 2006 American Thoracic Society doi: 10.1164/rccm.200511-1810OC
Physician-attributable Differences in Intensive Care Unit CostsA Single-Center StudyDivision of Pulmonary and Critical Care Medicine, Case Western Reserve University School of Medicine at MetroHealth Medical Center, Cleveland, Ohio Correspondence and requests for reprints should be addressed to Allan Garland, M.D., M.A., Division of Pulmonary and Critical Care Medicine, MetroHealth Medical Center, 2500 MetroHealth Drive, Cleveland, OH 44109. E-mail: agarland{at}metrohealth.org
Rationale: Variation in practice and outcomes, not explained by patient or illness characteristics, is common in health care, including in intensive care units (ICUs). Objective: To quantify within-ICU, between-physician variation in resource use in a single medical ICU. Methods: This was a prospective, noninterventional study in a medical ICU where nine intensivists provide care in 14-d rotations. Consecutive sample consisted of 1,184 initial patient admissions whose care was provided by a single intensivist. Multivariate models were constructed for average daily discretionary costs, ICU length of stay, and hospital mortality, adjusting for patient and illness characteristics, and workload. Measurements and Main Results: The identity of the intensivist was a significant predictor for average daily discretionary costs (p < 0.0001), but not ICU length of stay (p = 0.33) or hospital mortality (p = 0.83). The intensivists had more influence on costs than all other variables except the severity and type of acute illness. Average daily discretionary costs varied by 43% across the different intensivists, equating to a mean difference of $1,003 per admission between the highest and lowest terciles of intensivists. Conclusions: There are large differences among intensivists in the amount of resources they use to manage critically ill patients. Higher resource use was not associated with lower length of stay or mortality.
Key Words: costs and cost analysis • health resources • health services research • intensive care units
Widespread variation in practice and outcomes, not explained by patient or illness characteristics, is common in health care. Variation is found for many parameters in a broad range of settings (1–12), and is not limited to the United States (13–19). Large variation has also been noted in intensive care units (ICUs) (1, 7, 10, 16, 20–24). Such variation is important because it suggests that suboptimal care may be common. Substantial variation in patterns of care and outcomes has been attributed to individual physicians (4, 9, 13, 20, 25–28). However, we are not aware of any published reports that quantitatively address variation in ICU care attributable to individual intensivists. We hypothesized that there is important within-ICU, between-physician variation in the cost of ICU care that can be specifically attributed to the intensivists. We also hypothesized that intensivists cannot accurately judge the magnitude of resources they use in providing care. Some of the results of these studies have been previously reported in the form of an abstract (29).
We prospectively collected these data in the 13-bed closed-model medical ICU of a 520-bed county-owned, university-affiliated teaching hospital between July 2002 and March 2005, excluding March–May 2004 due to personnel limitations. The rotating ICU team comprises a board-certified intensivist, an ICU fellow, and five house officers who take overnight call on a 1-in-5 rotation. Nine intensivists assigned to 14-d intervals share ICU coverage. Neither the intensivist nor ICU fellow stays in the hospital overnight. More detailed information about the organization and functioning of this ICU is available in the online supplement. Multivariable linear or logistic models were constructed for average daily discretionary ICU costs, ICU length of stay (LOS), and hospital mortality. We log-transformed cost and LOS, and defined discretionary costs as including pharmacy, radiology (imaging and interventional), laboratories, blood bank, and echocardiography. We expressed costs obtained from the hospital's cost analysis system (Trendstar; McKesson Corporation, San Francisco, CA) as 2005 U.S. dollars. ICU LOS was measured in hours. Because transfer out of the ICU is often delayed due to limited ward bed availability, ICU LOS was defined as the interval from ICU admission until transfer was requested. To uniquely associate care with intensivists, we included patients only if the same intensivist was on duty each day of their ICU admission. To avoid erroneous mortality rates, we analyzed only initial admissions. The impact of the intensivists was assessed by including indicator variables in the models. Models included adjustments for demographics, comorbidities, type and severity of acute illness, invasive mechanical ventilation, the source of ICU admission, ICU workload, and any limitations placed on life-supporting therapies before ICU admission. Demographics were age, sex, and race dichotomized into white versus nonwhite individuals. Comorbidity was quantified as the number of comorbid conditions, as described by Elixhauser and colleagues (30). Acute diagnostic grouping was the organ system or illness responsible for ICU admission (respiratory, cardiovascular, gastrointestinal, neurologic, miscellaneous medical conditions, or surgical conditions including trauma). Severity of acute illness was measured as worst value in the initial 24 h of the Glasgow Coma Scale (GCS) score (31), and the APACHE II acute physiology score excluding its neurologic subcomponent (APS-N) (32). The source of ICU admission was the emergency department, hospital ward, another ICU, an outside hospital, or other sources. ICU workload was measured as the number of ICU admissions per day, and the ICU census averaged over each patient's ICU LOS. The cost model also included the actual time in the ICU as a covariate. Additional detail on modeling is provided in the online supplement. To evaluate the intensivists' ability to discern their practice styles, after reviewing the observed variation in discretionary costs, each intensivist completed a survey in which he or she guessed his or her position on this continuum using a visual analog scale. We assessed how accurately the intensivists judged their relative ranking of discretionary costs using a Spearman rank correlation. Data are presented as mean ± SD or as proportions. p values less than 0.05 are considered significant. Groups were compared using the t test, Fisher's exact test, or Hotelling's T2 test. Statistical analysis was performed using Stata 9.1 (StataCorp LP, College Station, TX). This study was approved by the hospital's institutional review board.
During the study period, there were 1,184 patients cared for by a single intensivist during their initial ICU admission. The mean number of patients cared for by the nine intensivists was 132 ± 67 (range, 32–253). These patients' characteristics are shown in Table 1. The respiratory system was the organ system most frequently responsible for ICU admission, comprising 27% of the patients. Three-quarters of admissions were from the emergency department. Slightly less than one-quarter of patients required invasive mechanical ventilation. Mean ICU LOS was 37.8 ± 30.3 h (range, 0.2–219.5 h). The ICU mortality rate was 7.3%, and another 5.3% survived the ICU but died before hospital discharge.
The model for average daily discretionary costs had an adjusted R2 = 0.255. All the covariates were significantly related to these costs except for race, the presence of an end-of-life order before ICU admission, and ICU workload. Sex was of borderline significance (p = 0.053). The identity of the intensivist was highly significant (p < 0.0001). To put these changes in perspective, Table 2 shows the relative predictive power of variables in the model. The influence of the intensivist on discretionary costs was less than that of the severity or type of acute illness, but greater than all other predictors, including comorbidities, demographics, the source admission to the ICU, and whether the patient required intubation.
To further illustrate the differences between intensivists, Table 3 shows the adjusted average daily discretionary costs for each intensivist in three different clinical scenarios: (1) a 40-yr-old white female with diabetes, admitted for respiratory failure and intubated in the emergency department with a GCS of 14, an APS-N of 11, who spent 5 d in the ICU; (2) a 68-yr-old African-American male with five major comorbid conditions, admitted to the ICU from an outside hospital for upper gastrointestinal hemorrhage due to hepatic cirrhosis, with a GCS of 9, an APS-N of 22, and a "Do Not Resuscitate" order prior to transfer, who died after 70 h in the ICU; and (3) a 20-yr-old white female with no comorbid conditions, admitted to the ICU from the emergency department for a suicide attempt via drug overdose, with a GCS of 15, an APS-N of 0, and who was in the ICU for less than 24 h. In these three scenarios, the mean adjusted average daily discretionary costs ranged $884 to $1,261, $1,578 to $2,250, and $407 to $581, respectively. Thus, discretionary ICU costs per day varied by 43% depending on which intensivist guided care.
To further investigate intensivists' spending, we compared the total discretionary costs incurred during each patient's entire ICU stay for the three highest spending intensivists with those of the three lowest spending ones. Table 4 shows that there is no predominant category of discretionary costs that distinguish higher from lower spenders.
We conducted sensitivity analyses to evaluate the robustness of our findings. We modeled the discretionary costs excluding intensivist I, who had the fewest number of patients in this dataset. This dataset showed virtually identical results, with an adjusted R2 = 0.245, and a p value less than 0.0001 for the intensivist's identity. We then modeled discretionary costs limited to the first ICU day in all 2,342 patients admitted during the study period. This model had an adjusted R2 = 0.142 and the intensivist's identity was statistically significant (p = 0.036). The ICU LOS model had an adjusted R2 = 0.136. Significant predictors were age, APS-N, source of ICU admission, intubation status, and ICU workload. The intensivist's identity was not significant (p = 0.325). For hospital mortality, significant predictors were age, APS-N, GCS, acute diagnostic group, comorbidity index, and the source of ICU admission. This model had good fit (Hosmer-Lemeshow test, p = 0.149), good discrimination (c-statistic = 0.891), and had adequate information content for the number of independent variables modeled. The intensivist's identity was not significant (p = 0.83). Figure 1 shows the results from the questionnaires administered to the nine intensivists. The rank correlation coefficient between their actual average daily discretionary costs and the perceptions of this parameter by the intensivists was 0.78 (p = 0.013).
Our analysis demonstrates large differences between ICU physicians in the amount of resources used to manage critically ill patients in a single medical ICU. Differences in the use of diagnostic and therapeutic interventions were not associated with mortality or ICU LOS. The effect of physicians on discretionary costs was large. The only influences that exceeded that of the intensivists were the severity and type of acute illness. For example, the intensivists' influence on daily discretionary costs was 2.4-fold larger than the effect of comorbid conditions, and threefold larger than that of patients' demographic characteristics (Table 2). These differences are economically relevant, with the top tercile of intensivists spending $1,003 more on discretionary costs over the average ICU admission compared with the bottom tercile of intensivists. Table 4 also shows that higher spending intensivists order more in each subcategory of discretionary resources compared with their lower spending colleagues. Such consistent differences in resource use, without concomitant differences in outcomes, allow for identification of differing "practice styles" among intensivists. Variation in health care has been extensively studied and is ubiquitous (1–12). Differences exist by geographic region (1–3), hospital (4–6), insurance status or system (7–12), and physician specialty (9). ICUs share this phenomenon (1, 7, 10, 16, 20–24). For example, the odds ratio across 34 ICUs for using pulmonary artery flotation catheters varied by 38% according to the patient's race and 33% according to insurance status, but by 200 to 400% according to how the ICU was organized and staffed (22). Prior studies in non-ICU settings showed that there is substantial variation in patterns of care and outcomes attributable to individual physicians or physician characteristics (4, 9, 13, 20, 25–28). After adjustment for severity of illness, the identity of the physician was associated with 40% variation in hospital charges and 57% variation in LOS for general medical ward patients (26). This 40% variation in hospital charges is very close to the 43% figure for discretionary ICU costs we observed. Also similar to our findings, other investigations have found that the amount of variation attributable to individual physicians is comparable to or greater than the amount due to patient and illness characteristics (4, 13, 25, 33). In a study on the costs of inpatient care related to giving birth, physicians accounted for approximately the same amount of variation as did patient characteristics, but physicians' characteristics (e.g., years in practice, board certification) accounted for just 10% of the influence of the physicians (25). This interesting result suggests that the differences among physicians in their practice styles transcend objective characteristics related to their training. The only prior work that quantified the effect of physicians on ICU care found that risk-adjusted mortality varied from 21 to 34% across five groups of physicians in a medical ICU, with risk-adjusted LOS varying by 3 d across these groups (20). Contrary to our secondary hypothesis, the survey data show that the intensivists have an accurate sense of the ICU costs they generate (Figure 1). Perhaps the higher spenders believe their practice style produces better outcomes. Alternatively, practice styles may derive from a complex interaction between training and personality traits, such as response to uncertainty (34–36). We are not aware of any prior studies that have compared physicians' self-assessments with objective measurement of their practice patterns or outcomes. There are a number of observations to make about our findings. First, the lack of influence of the intensivists on ICU LOS may reflect, at least in part, the reality that decisions to transfer patients out of this ICU are often influenced by bed availability within and outside the ICU. Support for this claim can be found in the observation that both ICU workload variables are significant predictors of ICU LOS. For example, an increase of one patient per day in the average ICU admission rate results in a 6.2-h decline in ICU LOS. Although in previous work from an ICU under less bed pressure we found that ICU LOS is physician dependent (20), the current findings indicate that LOS is not a reliable way to assay for variations in patterns of care; more robust measures reflecting daily resource use are superior. Second, we used rigorous model-building methods to ensure that artifacts, a broad range of potentially confounding factors, and erroneous assumption of linear relationships did not confuse the results (37). Third, although the models for ICU costs and LOS were highly significant, their overall predictive power, as indicated by the R2, were modest. Fourth, the models were limited to patients with only a single intensivist involved in their ICU care. This was done because the major goal of the study was to examine the influence of individual physicians on care. It would be difficult to clearly identify the role of the intensivists for any patient cared for by more than one of them. However, this restriction produced a cohort with characteristics different from the 1,158 ICU patients admitted during the study who were cared for by multiple intensivists. This is expected since the longer patients stay in ICU, the more likely they are to have more than one intensivist participating in their care. In particular, compared with patients cared for by a single intensivist, those with multiple intensivists had longer ICU LOS (114 ± 130 vs. 38 ± 30 h, p < 0.001), were sicker (APS-N, 13.6 ± 6.3 vs. 11.4 ± 6.0; p < 0.001), and had higher hospital mortality (16.8 vs. 12.6%, p = 0.005). They did not, however, differ in daily discretionary costs ($1,053 ± $1,022 vs. $1,084 ± $1,235, p = 0.50). Evidence that our findings are not dependent on special characteristics of the cohort studied comes from modeling discretionary costs from the initial ICU date for all 2,342 patients. In this model, differences among intensivists were still statistically significant (p = 0.036). Fifth, seasonal differences could potentially confound our findings. However, this is unlikely since there was no difference in the seasonal distribution of the ICU assignments among the nine intensivists (p = 0.86, Fisher's exact test). Sixth, the attending intensivists in this ICU function within an organizational structure that includes house officers at the resident and fellow levels. Because these other physicians did 4-wk rotations in the ICU, it is possible that their presence across the rotations of two different intensivists (who did 2-wk rotations) would partially obscure differences in care by the intensivists. Thus, variation attributable to intensivists might be even larger in ICUs without house staff. The major limitation of this study is that it evaluated a single ICU, raising concern about the generalizability of our results. However, as described in the online supplement, the nature, organization, operation, staffing, and case mix of our ICU is within the mainstream of large academic medical centers. Although other ICUs, which differ in type, organization, structure, case mix, geographic location, and so forth, may have less or more between-physician variation, we expect that all ICUs experience such variation to some degree. For example, there may be even greater variation in the ICUs of private, community, nonteaching hospitals that comprise the majority of all ICUs in the United States (38). ICUs in such institutions are often less structured, with less care provided by intensivists; care is provided by multiple groups of physicians with varied training instead of a cohesive team led by an intensivist from a single group; there is no in-hospital physician coverage after usual business hours; and there is an absence of house officers (38–40). Another concern relating to generalizability is that more comprehensive use of care protocols in other ICUs may reduce or eliminate the variation we observed in our unit. Although care pathways and other types of protocols are increasingly used, those currently in wide use apply to only a fraction of the conditions treated and situations encountered in ICUs. Moreover, most evidence indicates that the presence of clinical practice guidelines is not a very effective method of changing physicians' practice (41–43). Few protocols comprehensively guide use of the varied categories of costs we considered (e.g., imaging studies). Indeed, even patients whose care is highly protocolized, such as those undergoing coronary artery bypass surgery, experience substantial cost variability unrelated to their clinical characteristics (44, 45). In an attempt to discern whether our findings apply to a less heterogenous subset of patients, we modeled total discretionary ICU costs for our largest diagnostic subset, those with respiratory diagnoses. For this subset, the identity of the intensivist was still statistically significant (p = 0.017). In the end, further studies are required to evaluate the generalizability of our results. Although a multicenter study would have advantages regarding generalizability, the need to standardize for between-ICU variation presents complex challenges in assessing between-physician variation. Our single-center study allows for easier and potentially more valid comparisons between individual physicians. In this single-ICU study we showed large differences in ICU resource use that are attributable to differences in physicians' practice styles. Higher spending intensivists did not generate better outcomes than their lower spending colleagues. It appears possible to reduce ICU costs without worsening outcomes if we can alter physicians' practice styles. However, because we also found that higher spending intensivists were generally aware that they are higher spenders, getting them to change their practice style will require more than providing them with this information. Indeed, audit and feedback have limited utility in changing physician practice (46). Although more research is needed to understand the origin of practice styles and how to change them (47–49), we do not believe that efforts targeting individuals can solve the problem of variation in health care. Instead, we believe that variation, like medical errors, can most effectively be reduced by a paradigm shift where medicine evolves into a "culture" that explicitly promotes consistency of care across physicians (43, 50, 51).
This article has an online supplement, which is accessible from this issue's table of contents at www.atsjournals.org Originally Published in Press as DOI: 10.1164/rccm.200511-1810OC on September 14, 2006 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 November 25, 2005; accepted in final form September 11, 2006
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
AT A GLANCE COMMENTARY
, Spearman rank correlation coefficient.
