In this systematic review, we assessed the relationship between hospitalist physicians and the quality of inpatient care delivery. Forty-six of the sixty-five reviewed articles demonstrated that hospitalists delivered a higher quality of care to their patients compared to traditional inpatient physicians, and only seven studies indicated worse quality under the care of hospitalists. Superior outcomes were demonstrated across all care settings, regardless of study design, hospital type, patient eligibility, or physician practice structures. Stratifying these findings according to the area of quality examined showed improvements in operating efficiency among hospitalists (43 of 59 evaluations); however, there were few significant differences between physicians on process measures (15 of 26 evaluations) or clinical outcomes (33 of 52 evaluations). Taken together, our review of the current evidence suggests that hospitalists provide a level of clinical care that is comparable to that of their colleagues; however, their enhanced on-site availability and additional time spent on service suggests that the hospitalists' primary value likely comes from their ability to provide the same quality of clinical care in shorter periods of time, as evidenced by reductions in patients' average length of hospital stay reported in selected studies. Decreases in operating costs appear to be achieved largely by an increase in patient processing as opposed to reductions in the type and intensity of services provided. While there is no evidence to suggest that hospitalists provide a higher quality of clinical care, improvements in efficiency do not appear to come at the expense of clinical outcomes or patient and family satisfaction.
Despite these promising findings, many of the included studies had important methodological limitations, which decreases our confidence that findings reflect an accurate indication of hospitalist performance. Small sample sizes and inadequate statistical power were an issue in many studies, making it difficult to comment on whether hospitalists can decrease the incidence of rare outcomes such as in-hospital mortality or readmissions. The nonrandom allocation of patients frequently resulted in selection bias to preferred physician structures, where important covariates such as patient age, sex, ethnicity, insurance status, and preexisting comorbidities were often excluded from statistical models. Together, these factors resulted in poorly matched comparison groups and unadjusted biases. Finally, the statistical analyses used in selected studies were rarely conducted appropriately. Clinical indicators were frequently estimated among populations that were not actually at risk for the outcomes of interest, and inferences about quality were made at the level of providers without accounting for the clustering of patients within physicians. Furthermore, these methodological issues persist despite numerous calls urging researchers to enhance the rigor and reporting of the care provided by hospitalists compared with that offered by other healthcare providers.
Our findings are consistent with those reported in previous systematic reviews by Coffman and Rundall  and Landrigan et al.  suggesting improved performance by hospitalists based on the indicators of operating efficiency with no significant differences in patient outcomes between providers. These findings stand in contrast to those of Peterson's recent review , which found improvements in some process and outcome measures in addition to efficiency gains. It is worth noting that articles judged to be of 'poor' quality were excluded from Peterson's review, which may explain some of the deviations in our conclusions. When we attempted to replicate a version of Peterson's approach by excluding articles with quality scores below 50% (n = 35), we found little evidence to support processes or outcome improvements by hospitalists; however, 40% (n = 13) of the evaluations included in Peterson's review were found to have low quality scores in our review using the modified Downs and Black checklist .
In this systematic review, we propose a modified version of Donabedian's  framework as a simple conceptual map for understanding and synthesizing hospitalist performance, recognizing that an organization's structures, processes, and outcomes are interrelated and influence one another. By organizing these relationships into categories, researchers can logically predict and test relationships between constructs of interest and, in doing so, facilitate progression in the field of hospital medicine and quality initiatives. Structural differences between physician models should correlate with changes in the processes of care delivery, which in turn help drive improvements in operating efficiency and clinical outcomes. The results summarized in this review are important, as they suggest that the identification, labelling, and comparing of physicians as either 'hospitalists', 'traditional academic attending physicians', or 'community-based' providers is not sensitive enough to adequately differentiate the key structural characteristics which define hospitalists as distinct from other inpatient physicians and subsequently drive improvements in patient-level outcomes. The list of structural characteristics included in our conceptual model (Figure 1) quickly makes it apparent that inpatient physicians have access to many of the same resources and supports, regardless of job title, training, or time spent on service. By restricting all organizational aspects of a practice model to a single explanatory dummy variable as the vast majority of hospitalist evaluations have done, we do see evidence of improved performance in operating efficiency; however, we do not have a clear picture of where or how these efficiency gains occur and why we do not see similar improvement in related areas of quality (mainly processes and outcome measures).
Recognizing that hospitalists are now firmly entrenched within a large proportion of North American hospitals, if we wish to improve the quality of inpatient care delivery and introduce funding models that reward providers and/or institutions on the basis of their performance, further descriptive research labelling, categorizing, and analyzing of physicians according to their practice structures alone is unlikely to advance the research field in a way that will help inform organizational decision-making or health policy. Future research should instead shift toward developing better conceptual and theoretical models that identify and measure specific structural differences between physician practices, organizational issues that affect hospitalist groups, and the process mechanisms whereby hospitalist-based physicians have an increased opportunity to intervene.
On the basis of the findings of this review, we suggest that one of the key structural characteristics driving efficiency improvements among hospitalists is likely the increased time spent attending on the inpatient service and its subsequent impact on inpatient volume. Hundreds of articles published over the past three decades have shown that processes utilized and outcomes of care achieved are better among healthcare providers who perform them more frequently [6, 91]. These volume-outcome associations have been demonstrated across a wide range of study designs, patient populations, health delivery models, and outcomes examined, and they persist despite extensive adjustment for organizational differences between institutions. While the categorical classification of hospitalists implies a volume-outcome relationship, only three studies included in this review specifically examined case volume at the provider level as an explanatory variable of quality outcomes [53, 54, 92]. Many hospitalists choose to practice part-time. As such, the annual volume and experience of a part-time hospitalist may actually approach that of some comparative providers, potentially washing out any improvements in quality that may be driven by volume as opposed to the portion of a physician's practice which is dedicated to inpatient care delivery, a common approach used to define hospitalists. This effect was demonstrated by Lindenauer et al. , who found that hospital length of stay and costs varied by <0.10 days and $15, respectively, among providers in models that were not adjusted for physicians' annual case volume.
By examining the quality of general inpatient care as a function of a physician's annual case volume, we can also extend the application of this literature to other healthcare models around the world which have instituted parallel inpatient practices without necessarily establishing formalized hospitalist programs. For example, inpatient care delivery in Australia, New Zealand, the United Kingdom, Singapore, and several other former British colonies is similar to the North American hospitalist model in that primary care is handed over to a separate system of specialists and consultants (most often general internists and/or general surgeons) once a patient is admitted. Like the hospitalist, the specialist then 'owns' the patient for the duration of hospitalization, providing the majority of their clinical services within the hospital setting. In this manner, several structural characteristics of the hospitalist and specialist models overlap: Both have high annual inpatient volume, which theoretically enhances clinical expertise and improves patient outcomes, and both operate in a routine environment where familiarity with staff, services, and technological resources support efficient practice. There are, however, a few key differences. Hospitalists tend to practice using a team-based approach where patients, call hours, and vacation time are rotated according to prearranged contracts, while specialists still tend to operate individually, negotiating their work hours directly with the hospital administration. Furthermore, inpatient specialists frequently hold higher levels of medical certification than many North American hospitalists, especially in Canada, where more than 90% of hospitalists hold only a general medical licence and no formalized training in hospital medicine . Finally, there is the issue of incentives. Financial and other incentives for improving quality and efficiency are more common for hospitalists in the United States, while inpatient care in other countries is traditionally publicly funded. As a result, the need for these providers to modify their performance is frequently generated by negative pressure to reduce inefficiencies, potentially offsetting any intrinsic motivation to provide better care.
Interestingly, none of the hospitalist evaluations published to date have examined process indicators relating to the timeliness of care delivery, which would theoretically drive efficiency gains within our conceptual framework. In addition, transitions of care and communication patterns among hospitalists, patients, and their primary care physicians remain virtually unexplored and are important areas for further work. While computationally complex, this review highlights the need for multilevel, multisite studies which integrate the organizational effects of hospitals with more complete and informative data on the structure of hospitalist programs when undertaking evaluations of provider performance. Superior statistical models need to be used that control for patient, physician, and hospital-level confounding to understand whether higher inpatient quality reflects better hospital staffing and/or administration, organizational cultures that support hospitalist groups, or true improvements in the processes of care delivery by hospitalist physicians. Finally, the general quality of reporting in published studies can be improved by stating source populations, any inclusion versus exclusion criteria, patient and physician sample sizes within each comparison arm, and the number of patients lost to follow-up or excluded because of missing data. Disclosure of any performance incentives and funding sources, as well as the role of additional healthcare providers, should also be encouraged.
Strengths and weaknesses
To our knowledge, this is the most comprehensive review of hospitalist performance conducted to date. While formal registration of the review was not undertaken, extensive attempts were made to prevent review-level bias, and the design, population, research questions, and literature search methods were all specified a priori according to the Participants, Interventions, Comparisons and Outcomes, or PICO, method  as well as the PRISMA guidelines . We included studies of all methodological quality levels, with no restrictions on publication language, inpatient populations, physician practice structures, or outcomes examined. In addition, this is the first systematic review to assess the methodological quality of the hospitalist literature in which an objective checklist was employed that has been validated for use in both experimental and observational research . We tested the sensitivity of our findings to methodological quality, demonstrating that our conclusions are supported in both high and low quality studies, but highlighted that poor quality studies were more likely to report better performance among hospitalists, a result which may have been driven largely by confounding. Finally, we have developed and presented a conceptual framework for synthesizing and evaluating hospitalist performance. By situating our conclusions within this underlying framework, we were able to identify several gaps in the evidence where hospitalist performance appeared to deviate from its theoretical foundation. We have highlighted key areas of interest that hospitalist researchers may wish to explore in the coming years.
Despite these strengths, several weaknesses in our review should be noted. Given the heterogeneity of designs and outcomes examined among studies, we were unable to conduct formal meta-analyses or generate summary estimates of risk for any of the outcome measures. While meta-analysis would be powerful for estimating the overall impact of hospitalists on the effectiveness and efficiency of inpatient care delivery, the validity of this approach rests largely on the quality of reporting in the original studies, and 53% of the reviewed studies did not report enough information to compute standard effect sizes and/or margins of error. The pooling of results is also considered inappropriate when unadjusted biases are suspected. Despite this limitation, decreases in the length of stay and the cost of care were demonstrated across all practice settings and patient populations, strongly suggesting that hospitalists do improve the efficiency of care delivery. Assessing the methodological quality of individual studies is widely accepted as good practice in systematic reviews of randomized, controlled trials; however, the use of quality assessment tools to appraise observational studies is less established. We used a validated and reliable checklist that has demonstrated high internal consistency and good test-retest and interrater reliability for both randomized and nonrandomized studies . Nonetheless, each study is unique, and we recognize that a quality checklist may not include all items that are relevant for a particular topic and may include some items that are irrelevant, which can result in the misclassification of a study's quality. We attempted to minimize this risk by modifying the original Downs and Black checklist  to include several items specific to reporting within hospitalist comparisons and to remove one question that was not applicable to these designs. One author (HLW) extracted data from the selected publications which could introduce errors in our analyses; however, in those instances where required information was unclear, input was sought and consensus was reached between both authors. Finally, the majority of studies included in this review did not adjust for important confounders of quality such as patient age, sex, insurance status, comorbidities, and hospital and physician clustering. Recognizing that risk adjustment can have a profound impact on individual study results, any conclusions drawn from a systematic review of hospitalists' performance may change substantially, depending on the type of risk adjustment employed and on inclusion versus exclusion criteria. The trends identified in this review should be verified and reevaluated in the coming years as the methodological quality of new evaluations continues to improve.