There is growing interest in collaborative management of surgical patients. However, few data describe how medical consultation influences quality of care or resource use. The objective of this study was to determine whether medical consultation improves care in surgical patients.
Observational cohort of patients undergoing surgery between May 1, 2004, and May 31, 2006, at a university-based hospital. The outcomes included costs, hospital length of stay, use of preventive therapies (such as perioperative β-blockers) and clinical outcomes.
Of 1282 patients, 117 (9.1%) underwent a perioperative medical consultation. Consulted patients were of a similar age, sex, and race, but more frequently had an American Society of Anesthesiologists score of 4 or higher (34.2% vs 13.0%; P < .001), diabetes mellitus (29.1% vs 16.1%; P < .001), vascular disease (35.0% vs 10.6%; P < .01), or chronic renal failure (23.9% vs 5.6%; P < .001). After adjusting for severity of illness and likelihood of receiving a consultation, patients were just as likely to have a serum glucose level of less than 200 mg/dL (<11.1 mmol/L), receive perioperative β-blockers, or receive venous thromboembolism prophylaxis. Consulted patients had a longer adjusted length of stay (12.98% longer; 95% confidence interval, 1.61%-25.61%) and higher adjusted costs (24.36% higher; 95% confidence interval, 13.54%-36.34%). Patients who had a consultation from a generalist did not receive different quality of care, but had costs and length of stay similar to nonconsulted patients. Our results may be influenced by unaccounted referral bias or severity of illness.
Perioperative internal medicine consultation produces inconsistent effects on efficiency and quality of care in surgical patients. Modifying the consultative model may represent an opportunity to improve care.
Consultation is a core feature of inpatient internal medicine practice, and published guidelines for effective consultations1-3 seek to improve adherence to consultants' recommendations.4-6
In nonsurgical settings, consultation may facilitate higher-quality care for patients with acute myocardial infarction.7 The availability of a dedicated diabetes mellitus consultation service may shorten the hospital length of stay.8 In contrast, radiologic consultations intended to guide appropriate test use have little effect.9
Little is known about the effects of internist consultation on the care of surgical patients. Preoperative cardiologist10 or intensivist11 consultations focused on detection of unstable clinical disease, rather than on improving care processes more generally. One randomized trial12 of preoperative internal medicine consultation showed little effect on postoperative outcomes and did not study the quality of care. Medical-surgical “comanagement” attempts to improve on traditional consultation by increasing the availability of internists—but its effects on outcomes are modest at best—and effects on the quality of care are unknown.13-15
To understand the effect of perioperative consultation on patients undergoing major surgery, we analyzed data from an observational study at our academic medical center. We hypothesized that internal medicine consultation on the day before, day of, or day after surgery would increase the likelihood that a patient would receive therapies, such as perioperative β-blockade (PBB), prevention of venous thromboembolism, or tighter glucose control. In addition, we hypothesized that medical consultation would lead to reduced costs and length of stay.
Our study was conducted at the University of California, San Francisco (UCSF), Medical Center, composed of Moffitt-Long Hospital (a 400-bed center) and UCSF–Mount Zion Hospital (a 200-bed facility), located in San Francisco. Both hospitals have dedicated anesthesiologist-staffed preoperative clinics that evaluate patients 1 to 2 days before elective surgery. Although each site has different surgical services, surgical house officers rotate between hospitals; there are no differences in availability of nursing or ancillary staff. Patients undergoing elective surgery are screened in the Anesthesia Preoperative Clinic, which has no protocols for requesting perioperative medical consultation; those admitted for emergency or urgent surgery receive consultation at the discretion of the surgical team.
Internal medicine consultative services at UCSF are staffed by an attending physician and a fellow (for subspecialty services, such as cardiology) or a third-year internal medicine resident. Consultative services are available 7 days a week, 24 hours a day, at both sites, with fellows or house staff providing overnight and weekend/holiday coverage. For billing purposes, all initial consultations are required to state a reason for consultation using a standard template, but consultation templates contain no prompts for clinical care.
During the study, UCSF was actively seeking to increase use of PBB, venous thromboembolism prevention (VTEP), and tighter glucose control via a variety of educational and audit/feedback efforts targeting all clinical staff involved with surgical patients. Preprinted orders with criteria for selection of eligible patients were available in all clinical care areas. In fact, internal medicine physicians championed these guidelines.
Data were collected as part of the UCSF Perioperative Quality and Safety Initiative, an observational study of patients admitted for major surgery at UCSF Medical Center between May 1, 2004, and May 31, 2006. Patients were eligible if they were older than 18 years and if their medical record was selected for abstraction according to Center for Medicare and Medicaid Services criteria for public reporting of data regarding adherence to surgical site infection processes. This process randomly selects medical records of patients who underwent colon surgery, cardiac bypass or valve procedures, hip or knee arthroplasty, hysterectomy, or vascular surgery. The UCSF Committee on Human Research approved the Perioperative Quality and Safety Initiative.
Data were obtained from medical record abstraction and administrative sources. Trained medical record abstractors collected preoperative and postoperative clinical and medication data (particularly processes of care, such as VTEP) and information about the hospital course (such as complications). Medications were counted as administered only if noted as such in the medication administration record. Medical record reviewers were trained by one of us (A.D.A.), who directly observed the first 5 medical record abstractions; a random subset of 5% was reviewed afterward to ensure data validity.
Consultation data were collected from initial consultation notes from internal medicine services. Initial notes were used to collect data regarding the consulted service, reasons for consultation, and initial recommendations. To determine consultants' final recommendations, we followed up until consultants' notes stopped appearing in the medical record or the patient was discharged.
Consultations on the day before surgery (day − 1), the day of surgery (day 0), or the first postoperative day (day 1) were considered “perioperative consultations.” Patients who received consultation on other days and patients who received consultation from non–internal medicine services (eg, pain service) were not considered to have had a perioperative consultation. The UCSF billing systems provided information on patient socioeconomic data, costs, length of stay, and attending surgeon.
Definition of target populations and quality metrics
We defined patients eligible for VTEP using criteria from the American College of Chest Physicians.16 Patients eligible for VTEP were not ambulating by postoperative day 1 or 2 and did not have risks for adverse outcomes (eg, history of intracerebral hemorrhage) or a new VTEP contraindication. Within eligible patients, we defined our VTEP quality-of-care measures as receipt of oral warfarin sodium, subcutaneous low-molecular-weight heparin sodium, or subcutaneous unfractionated heparin on postoperative day 1 (most stringent measure of quality) or 2 (less stringent measure of quality).
Eligibility for PBB was defined using 2 or more minor predictors (ie, age > 65 years, hypercholesterolemia, smoking, or hypertension) or 1 major predictor (ie, history of coronary artery disease, transient ischemic attack, or stroke syndrome; creatinine level of ≥ 2.0 mg/dL [≥ 177 μmol/L]; chronic renal insufficiency; hemiplegia; claudication; diabetes mellitus; or plan for aortic aneurysm surgery), according to published recommendations considered to be highly evidence based when data were collected.17,18 In addition, patients receiving β-blockers as outpatients were eligible. Patients were ineligible for PBB if they had contraindications to β-blockade (such as bradycardia) noted at any time. All patients with a preoperative history of diabetes mellitus were eligible for tight glucose control. Perioperative β-blockade was defined as any use of β-blockers or α2-agonists (such as clonidine), also on postoperative day 1 or 2.
Glucose control quality-of-care metrics were defined using our site's standard of care at the time (all serum glucose levels < 200 mg/dL [< 11.1 mmol/L] on postoperative day 1 or 2).
We first compared patients who received consultation with those who did not using the Fisher exact test and Wilcoxon rank sum test for categorical and continuous data, respectively.
Using backward stepwise selection techniques along with manually entered variables, we fit multivariate models to determine associations between perioperative consultation and risks for complications after postoperative day 2, and total costs and length of stay. Items were selected based on an unadjusted association with the outcome of interest of P < .05, based on observed confounding with other independent variables, or to maintain face validity of the model. Cost and length of stay data were log transformed to normalize data and stabilize variance of residuals in multivariate models, with percentage differences attributable to consultation calculated using the following simple equation: 100 × (eβ − 1).
Patients did not receive consultation at random. Therefore, multivariate models were used to adjust for likelihood of receiving a consultation using a propensity score calculated in a logistic regression model with whether a consultation took place as the dependent variable. The propensity score was constructed using similar manual and automated methods as the core multivariate models. However, we selected variables associated with consultation at the P ≤ .20 level, according to standard methods.19,20 The propensity score model had good discriminative power (C statistic, 0.79) and was not overfit to our data (Hosmer-Lemeshow test, P = .60). All analyses were performed using SAS statistical software, version 9.1, for Windows (SAS Institute Inc, Cary, North Carolina).
Of 1282 patients, a surprisingly small proportion (9.1%) had consultation the day before, day of, or day after surgery. Patients who had perioperative consultation were of similar age, sex, and race, but less often underwent elective surgery. Patients who underwent consultation more often had coronary artery disease, hypertension, diabetes mellitus, atrial fibrillation, vascular disease, renal failure, congestive heart failure, or chronic obstructive pulmonary disease (P < .05 for all); a higher American Society of Anesthesiologists score (34.2% vs 13.0% had an American Society of Anesthesiologists score of ≥4; P < .001); and higher cardiac risk, as estimated by the Revised Cardiac Risk Index (17.9% vs 3.7% had an index of ≥3) (Table 1). Patients who received consultation also had longer surgical procedures than those who did not. In unadjusted comparisons, patients who had perioperative consultation were more likely to experience complications after surgery.
Within the 117 patients, 125 consultations were obtained. Of these 125 consultations, 66 were on the day after surgery, 38 were on the day of surgery, and 21 were on the day before surgery. The most common reason for consultation was for a specific medical problem, and cardiology was most frequently consulted (Table 2). The most common initial recommendation was to add a medication, but 17 patients had no clear recommendation. Few consultants specified that they were “signing off” or that they had arranged postdischarge follow-up.
Effects of consultation on care processes and complications
Associations between quality measures and consultation were variable in unadjusted analyses; after adjustment, no consistent trends toward higher-quality care among patients who underwent consultation emerged. For example, patients who underwent consultation had higher unadjusted odds for receipt of perioperative β-blockers on postoperative day 1; after adjustment, this difference remained nonsignificant (Table 3). For other measures, such as glucose control or VTEP, no significant associations were observed. Finally, patients who underwent consultation were more likely to have complications noted after postoperative day 2 in unadjusted and adjusted analyses.
Effects of consultation on length of stay and cost
Patients who received consultation had unadjusted cost (median [interquartile range], $155 020 [$101 473-$292 951] vs $74 237 [$53 824-$126 927]) (difference, 115.55%; 95% confidence interval, 87.57%-147.69%) and length of stay (median [interquartile range], 10 [7-18] days vs 6 [4-9] days) (difference, 87.20%; 95% confidence interval, 63.23%-114.68%) almost double that of patients who did not receive consultation. After adjustment, these differences were much less striking, although statistically significant (cost data difference, 24.36% [95% confidence interval, 13.54%-36.34%] and length of stay data difference, 12.98% [95% confidence interval, 1.61%-25.61%]). (Differences were calculated using the equation given in the “Analysis” subsection of the “Methods” section.)
Effects of specialist and generalist consultation
To understand whether a specialist's potentially narrower focus would influence the effect of consultation, we undertook subset analyses comparing no consultation with specialty and generalist consultation groups. Compared with patients who had specialty consultations, patients seen by generalists had a similar length of stay and no difference in the odds of receiving perioperative β-blockers (Table 4). However, patients seen by generalists were more likely to receive venous thromboembolism prophylaxis and had lower overall costs than specialists' patients. In comparisons of consultants' patients with those with no consultation, specialists' and generalists' patients had no difference in the odds for receiving PBB. Generalist patients had higher odds for venous thromboembolism prophylaxis and no difference in costs or length of stay; specialist patients had substantially higher costs and length of stay when compared with patients without consultation.
In this observational study, patients who received consultation, not surprisingly, had a higher likelihood of complications, longer length of stay, and higher costs, which persisted despite accounting for referral biases and severity of illness using all available data. However, consultation had no effects on quality of care for tightly defined subsets of patients in which all patients should have been eligible and the influence of residual confounding less pronounced. Although confounding likely contributed to persistently elevated risk of complications and higher resource use, our quality measures suggest that medical consultants may be missing an opportunity to improve care of surgical patients.
Despite use of all available data in multivariate models, it seems likely that our results were influenced by the patterns in which consultations took place and biases related to unmeasured elements of patient's history or illness. We attempted to account for referral bias with the propensity score method.19,20 In nonrandomized trials, propensity scores have strong face validity,21 and we have adhered to recommendations about how these scores should be constructed and used.22,23 More important, our methods for defining target populations for quality measures are in themselves a form of severity adjustment, making these outcomes the least vulnerable to biases related to illness severity.
Other than methodological concerns, there are a few potential explanations for why we did not see improvements in the quality of care or outcomes. Consultants at our site may have tended to focus only on the question at hand and “respecting thy turf”1 because recommendations focusing solely on the question may have a higher likelihood of being followed2,5; attenuation of differences and the suggestion of trends toward improved quality and less resource use in the patients receiving generalist consultations suggest this as a potential mechanism. It is also possible that consultants' recommendations were not followed even if the correct recommendation was made but not followed. If so, this disconnect represents an additional way to improve consultation. Finally, preprinted orders available for management of VTEP, PBB, and diabetes mellitus may have been viewed as being the purview of the surgical teams, rather than a broader responsibility.
Consultation in itself seems unlikely to precipitate complications, but consultants may have been better at detecting or documenting adverse events. Alternatively, consultation may have been prompted because the surgical team had concerns about a complication (or recognized one had taken place) or because they suspected a complication and were asking a consultant to aid diagnosis. These effects would have made it difficult to detect whether consultation actually reduced the risk for complications. Similarly, consultation may have led to higher costs and length of stay because consultants may have ordered tests or procedures.24 However, after adjustment, this difference was markedly smaller, suggesting that consultants may have been able to expedite other elements of care even as they prompted use of additional resources or that full adjustment for unaccounted biases would have narrowed (or reversed) higher costs and risks for adverse events in the consultation group.
As our study demonstrates, observational study of perioperative consultation poses a number of daunting challenges. Despite the importance of this area of clinical practice, randomized trials of medical consultation are scarce, a trend likely related to the complexity and practicality of carrying out such a trial. Few surgeons or patients are likely to agree to be randomized to receiving no consultation at the time the decision for a consultation is made. However, studies that randomize patients to targeted preoperative consultation and/or different models of management in the immediate perioperative period may be more feasible, and even desirable, given the difficulties with observational studies such as ours. Further study of comanagement models—a robust consultative model in which order writing and clinical assessment duties are shared between internists and surgeons—is also needed. In general, literature describing comanagement's effects has shown disappointingly modest benefits13-15 at higher incremental costs (in terms of supporting physicians); this small marginal benefit would argue strongly for studies that assign only higher-risk patients to comanagement systems or target comanagement for a shorter period around surgery. These studies would need to be preceded by formative research identifying “high-risk” patients and the high-risk periods for surgical complications. Finally, studies of changes to consultation templates or consultation guidelines may help increase use of preventive therapies and overcome turf issues.
Our study has several limitations in addition to the biases related to referral and severity of illness previously discussed. Sample sizes in some care process outcomes were relatively small compared with our overall population, and power in those subgroups is limited. However, within these subgroups, event and adherence rates were high, enabling us to detect relatively small differences. Our study did not collect detailed data regarding whether consultants' recommendations were followed, nor do we have information about whether a consultation was suggested by a primary care or other outpatient physician or about whether preprinted forms were used by consultants. Finally, our findings were derived from a single academic teaching system and may not apply to nonteaching centers where consultation systems and referral patterns may be different.
Medical consultation for patients undergoing surgery is a core element of internists' practice. With the growing complexity of perioperative care and pressures to improve (and publicly report) quality metrics, the need to maximize quality of care of surgical patients is increasing. Although the culture that prompts internist-surgeon consultation may not change overnight, increasing the focus of consultants and surgeons on specific care practices may catalyze a shift toward a culture that encourages, systematizes, or cues quality-of-care practices regardless of specialty. These care systems are worthy subjects for future study.
Correspondence: Andrew D. Auerbach, MD, MPH, Department of Medicine, University of California, San Francisco, Campus Box 0131, San Francisco, CA 94143-0131 (firstname.lastname@example.org).
Accepted for Publication: July 13, 2007.
Author Contributions:Study concept and design: Auerbach. Acquisition of data: Auerbach, Rasic, Sehgal, Ide, and Stone. Analysis and interpretation of data: Auerbach, Ide, and Maselli. Drafting of the manuscript: Auerbach, Rasic, and Maselli. Critical revision of the manuscript for important intellectual content: Auerbach, Sehgal, Ide, and Stone. Statistical analysis: Auerbach and Maselli. Obtained funding: Auerbach and Ide. Administrative, technical, and material support: Auerbach, Rasic, Sehgal, Ide, and Stone. Study supervision: Auerbach.
Financial Disclosure: None reported.
Funding/Support: This study was supported by the University of California, San Francisco, Perioperative Safety and Quality Initiative; research and training grant K08 HS11416-02 from the Agency for Healthcare Research and Quality (Dr Auerbach); and grant-in-aid 0455008Y from the American Heart Association (Dr Auerbach).
Role of the Sponsor: The funding bodies had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
et al. Why don't they listen? adherence to recommendations of infectious disease consultations. Clin Infect Dis
1212- 1218PubMedGoogle ScholarCrossref
HJ A randomized, controlled clinical trial of a geriatric consultation team: compliance with recommendations. JAMA
2617- 2621PubMedGoogle ScholarCrossref
ME Compliance with the recommendations of medical consultants. J Gen Intern Med
220- 224PubMedGoogle ScholarCrossref
ME The effectiveness of a consultation: compliance with initial recommendations. Am J Med
870- 876PubMedGoogle ScholarCrossref
et al. Consultation between cardiologists and generalists in the management of acute myocardial infarction: implications for quality of care. Arch Intern Med
1778- 1783PubMedGoogle ScholarCrossref
B Impact of endocrine and diabetes team consultation on hospital length of stay for patients with diabetes. Am J Med
22- 28PubMedGoogle ScholarCrossref
SJ Effect of mandatory radiology consultation on inpatient imaging use: a randomized controlled trial. JAMA
1595- 1598PubMedGoogle ScholarCrossref
TKC The value to the anesthesia-surgical care team of the preoperative cardiac consultation. J Cardiothorac Anesth
682- 687PubMedGoogle ScholarCrossref
M Preoperative intensive care unit consultations: accurate and effective. Crit Care Med
234- 239PubMedGoogle ScholarCrossref
RP Outpatient internal medicine preoperative evaluation: a randomized clinical trial. Med Care
498- 507PubMedGoogle ScholarCrossref
et al. Hospitalist-Orthopedic Team Trial Investigators, Medical and surgical comanagement after elective hip and knee arthroplasty: a randomized, controlled trial. Ann Intern Med
28- 38PubMedGoogle ScholarCrossref
H An internist joins the surgery service: does comanagement make a difference? J Gen Intern Med
440- 444PubMedGoogle ScholarCrossref
et al. Effects of a hospitalist model on elderly patients with hip fracture. Arch Intern Med
796- 801PubMedGoogle ScholarCrossref
et al. Prevention of venous thromboembolism: the Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest
338S- 400SPubMedGoogle ScholarCrossref
L β-Blockers and reduction of cardiac events in noncardiac surgery: scientific review. JAMA
1435- 1444PubMedGoogle Scholar
D The central role of the propensity score in observational studies for causal effects. Biometrika
41- 55Google ScholarCrossref
DB Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc
1984;79516- 524Google ScholarCrossref
PC Propensity score methods gave similar results to traditional regression modeling in observational studies: a systematic review. J Clin Epidemiol
550- 559PubMedGoogle ScholarCrossref
V Principles for modeling propensity scores in medical research: a systematic literature review. Pharmacoepidemiol Drug Saf
841- 853PubMedGoogle ScholarCrossref
et al. International Stroke Trial Collaborative Group; European Carotid Surgery Trial Collaborative Group, Evaluating non-randomised intervention studies. Health Technol Assess
iii- x, 1-173PubMedGoogle Scholar
MT Comparing generalist and specialty care: discrepancies, deficiencies, and excesses. Arch Intern Med
1596- 1608PubMedGoogle ScholarCrossref