Patient, Surgeon, or Hospital: Explaining Variation in Outcomes after Colectomy

Abstract

BACKGROUND:

Complication rates after colectomy remain high. Previous work has failed to establish the relative contribution of patient comorbidities, surgeon performance, and hospital systems in the development of complications after elective colectomy.

STUDY DESIGN:

We identified all patients undergoing elective colectomy between 2012 and 2018 at hospitals participating in the Michigan Surgical Quality Collaborative. The primary outcome was development of a postoperative complication. We used risk- and reliability-adjusted generalized linear mixed models to estimate the degree to which variance in patient-, surgeon-, and hospital-level factors contribute to complications.

RESULTS:

A total of 15,755 patients were included in the study. The mean hospital-level complication rate was 15.8% (range, 8.7% to 30.2%). The proportion of variance attributable to the patient level was 35.0%, 2.4% was attributable to the surgeon level, and 1.8% was attributable to the hospital level. The predicted probability of complication for the least comorbid patient was 1.5% (CI 0.7–3.1%) at the highest performing hospital with the highest performing surgeon, and 6.6% (CI 3.2–12.2%) at the lowest performing hospital with the lowest performing surgeon. By contrast, the most comorbid patient in the cohort had a 66.3% (CI 39.5–85.6%) or 89.4% (CI 73.7–96.2%) risk of complication.

CONCLUSIONS:

This study demonstrated that variance from measured factors at the patient level contributed more than 8-fold more to the development of complications after colectomy compared with variance at the surgeon and hospital level, highlighting the impact of patient comorbidities on postoperative outcomes. These results underscore the importance of initiatives that optimize patient foundational health to improve surgical care.

Surgical quality improvement initiatives often address care as it is delivered by the surgeon or hospital system. Perioperative care has certainly improved through these initiatives, and care delivered to patients undergoing colectomy has received particular attention given the ubiquity of the operation. For example, a 3-item bundle has been successful in decreasing rates of postcolectomy surgical site infections from 6.7% to 3.9%.¹ Moreover, there has been considerable focus on enhanced recovery protocols (ERPs) to decrease length of stay and postoperative complication rates.^2–7 Interestingly, many of the studies evaluating benefits of ERPs lack generalizability,^8,9 and a recent study showed that real-world effects of ERPs are likely lower than originally reported at specialty centers, such that complication rates decreased from 16.9% to 14.6%, and length of stay decreased by about one half of a day.¹⁰

Considering the depth of these interventions, observed improvements in postoperative outcomes are overall quite modest and postcolectomy complication rates remain high.¹⁰ Although this is certainly multifactorial and contingent on a combination of factors at the patient, surgeon, and hospital levels, the relative contribution of patient characteristics, surgeon performance, and hospital setting to the development of postoperative complications has not been established. Previous work evaluating a statewide quality improvement collaborative suggested that variance at the surgeon and hospital level contribute minimally, but this study included a wide variety of cases in which overall complication rates were low.¹¹ A better understanding of the drivers of postcolectomy complications is necessary in elucidating more effective targets for care improvement.

In this context, we used a detailed regional data set to understand the degree to which postoperative outcomes in colectomy are attributable to variation at the patient, surgeon, and hospital level. We hypothesized that the contribution of patient comorbidity is substantially higher than that of surgeon- and hospital-level factors in the development of postoperative complications.

METHODS

Data source and study population

This was a retrospective cohort study of patients in the Michigan Surgical Quality Collaborative (MSQC). The MSQC is a statewide Collaborative Quality Initiative (CQI) founded in 2005 with 72 member hospitals representing all the hospitals that perform major surgery across the state of Michigan. MSQC receives full financial support from Blue Cross Blue Shield of Michigan and Blue Care Network. Data are prospectively collected from a sampling of surgical cases using specific methodology to minimize selection bias.¹² Data collection is complete in 95% of cases and occurs at the hospital level by trained data abstractors. Data are abstracted by trained surgical clinical quality reviewers who have access to patients’ entire medical record and include patient characteristics, perioperative processes of care, and 30-day postoperative outcomes from general, vascular, and gynecologic surgical procedures. Using this dataset, we identified all patients 18 years of age or older who underwent laparoscopic or open elective colectomy between 2012 and 2018. Laparoscopic and open colectomies were identified using current procedural terminology (CPT) codes (Supplemental Digital Content 1, available at http://links.lww.com/XCS/A49). Exclusion criteria included pregnant women, because they are not included in MSQC data. This study was exempted from regulation by the IRB of the University of Michigan.

Outcomes

Our primary outcome of interest was a composite of all non–urinary tract infection (UTI) complications occurring up to 30 days postoperatively. Complications included in the composite measure were superficial, deep, and organ space surgical site infections (SSI), anastomotic leak, sepsis (including severe sepsis and septic shock), postoperative transfusion, pneumonia, deep vein thrombosis, pulmonary embolism, unplanned intubation, acute kidney injury, myocardial infarction (MI), stroke, and C. difficile infection. This was made into a binary outcome, such that patients were classified as having developed any complication within 30 days or no complication within 30 days.

Statistical analyses

Descriptive statistics were used to characterize the patient cohort. A hierarchical, mixed-effects, logistic regression model was estimated to identify patient characteristics independently associated with the development of a postoperative complication. For our model, we used 3 levels, with patients clustering within surgeons and surgeons clustering within hospitals. The first level contained patient covariates including all their demographics and comorbidities. The second level contained only the surgeon identifier, and the third level contained only the hospital identifier. The distribution of the random effects was assumed to be normal and the individually observed random effects were not correlated with the regressors (individual variables) in our model.¹³ Concordance statistics and calibration plots were generated to ensure appropriate model fit. Two-sided significance tests were used for all analyses, and significance was defined as p < 0.05. All statistical analyses were performed using SAS, version 9.4.

Risk and reliability adjustment

We calculated risk- and reliability-adjusted hospital complications rates using hierarchical logistic modeling with empirical Bayes estimates. This is a 2-step process that first calculates risk-adjusted hospital complication rates to account for differences in case mix between hospitals. In the second step, we adjust these hospital complication rates for reliability using empirical Bayes’ techniques as a means to adjust for statistical “noise” attributable to low case volumes. To do this, we used postestimation commands to calculate empirical Bayes’ estimates of each hospital random effect and added this random effect to the average hospital risk. We then performed an inverse logit to calculate the risk- and reliability-adjusted complication rate for each hospital. Previous work has shown that individual hospital-level risk assessment is subject to poor measurement reliability for hospitals that have low case volumes.^14,15 Using reliability adjustment shrinks observed complication rates toward the overall observed mean, and details regarding this technique have previously been described in detail.^15,16 In brief, this is done by weighting the observed rate based on the reliability of the measure, which in this case is only a function of the number cases for each hospital and attributing the remaining weight to the overall mean. In this way, the magnitude of shrinkage is smaller for hospitals with high case volumes.

Estimating proportional variation

We then estimated the variance partitioning coefficients (VPC) of the patient, surgeon, and hospital levels. The VPC ranges from 0 to 1 and is a proportion measuring the level-specific contribution towards the total outcome variance. For example, if the surgeon-level VPC is 0.05, that means that the proportion of total variance in complication rate attributable to the surgeon level is 5%. To calculate the VPCs, we first generated a null model that contained only surgeon-specific and hospital-specific random effect terms. We then generated a fully-fitted model that included fixed patient characteristics, and from this, we obtained the surgeon- and hospital-level VPC using postestimation commands. To obtain the patient-level VPC, we used previously published methods that considered residuals of a logistic distribution.^17,18

Sensitivity analyses were performed to determine whether the degree of level-specific contributions varied with severity of complication. We used the same analysis described above to determine the VPC of the patient, surgeon, and hospital levels toward the development of serious complications and minor complications. We defined serious complications as anastomotic leak, cardiac arrest, deep SSI, organ space SSI, MI, septic shock, sepsis, stroke, or unplanned intubation and minor complications as the remaining complications not included in serious complications.

Generating predicted probabilities

Using our previously described model, we generated a predicted probability of a postoperative outcome for each patient in our cohort using only their fixed characteristics. We then ranked patients by their predicted risk of complication to identify the least comorbid patient and the most comorbid patient in our sample. We then used surgeon- and hospital-level specific random effects to reflect different care settings and predicted the probability of a complication for each patient in the following scenarios: 1) care from the highest performing surgeon at the highest performing hospital, and 2) care from the lowest performing surgeon and the lowest performing hospital. To identify the highest performing surgeons and hospitals, we identified the smallest surgeon-specific and hospital-specific random effects based on their log odds of complications. To identify the lowest performing surgeons and hospitals, we identified the largest surgeon-specific and hospital-specific random effects based on their log odds of complications.

RESULTS

Demographics

In total, 15,755 patients meeting inclusion criteria underwent elective colectomy during the study period (Table 1). Most colectomies were performed using minimally invasive techniques (67.6% minimally invasive; 32.4% open). Patient sex was nearly evenly split (53.6% female; 46.4% male). Most patients (90.8%) were above age 45. The majority of patients identified as white (87.9%). Comorbidities were common; the most common of these were hypertension (54.6%), obesity (39.1%), cigarette use (21.0%), sleep apnea (20.0%), and diabetes (17.2%). Adjusted odds ratios (aOR) for common comorbidities are listed in Table 2. A few factors associated with a higher odds of a postoperative complication included the following: older age (aOR 1.01; CI 1.00–1.01), female sex (aOR 1.16; CI 1.05– 1.28), wound classification of “contaminated” or “dirty/infected” (aOR 1.21; CI 1.07–1.37), and an open surgical approach (aOR 1.90; CI 1.71–2.11) as well as several patient comorbidities and a “dependent” functional status (aOR 1.72; CI 1.34–2.23).

Table 1.

Patient Demographics and Clinical Characteristics

Characteristic	Overall	No complication	Complication	p Value*
Age group, y, mean (SD)	62.7 (13.6)	62.3 (13.4)	64.7 (14.5)	<0.0001
Age group, n (%)				<0.001
<45 y	1,448 (9.2)	1,241 (9.3)	207 (8.4)
45–64 y	6,898 (43.8)	5,990 (45.1)	908 (36.9)
65+ y	7,409 (47.0)	6,060 (45.6)	1,349 (54.7)
Sex, n (%)				<0.001
Female	8,442 (53.6)	7,034 (52.9)	1,408 (57.1)
Male	7,313 (46.4)	6,257 (47.1)	1,056 (42.9)
Race, n (%)				0.001
White	13,413 (87.9)	11,357 (88.3)	2,056 (85.9)
Black	1,658 (10.9)	1,347 (10.5)	311 (13.0)
Other	177 (1.2)	151 (1.2)	26 (1.1)
Insurance, n (%)				<0.001
Other	10,716 (67.8)	8,849 (66.6)	1,867 (75.8)
Commercial insurance (non-HMO)	5,093 (32.2)	4,442 (33.4)	597 (24.2)
Weight, n (%)				0.006
Underweight	374 (2.4)	306 (2.3)	68 (2.8)
Normal weight	3,991 (25.4)	3,342 (25.2)	649 (26.4)
Overweight	5,213 (33.1)	4,470 (33.7)	743 (30.2)
Obese	6,150 (39.1)	5,150 (38.8)	1,000 (40.7)
Cigarette use,† n (%)				0.689
Yes	3,315 (21.0)	2,804 (21.1)	511 (20.7)
No	12,440 (79.0)	10,487 (78.9)	1,953 (79.3)
ASA classification, n (%)				<0.001
ASA 1	214 (1.4)	197 (1.5)	17 (0.7)
ASA 2	6,848 (43.5)	6,071 (45.7)	777 (31.5)
ASA 3	8,065 (51.2)	6,581 (49.5)	1,484 (60.2)
ASA 4 or 5	619 (3.9)	434 (3.3)	185 (7.5)
Ascites, n (%)				0.004
Yes	67 (0.4)	48 (0.4)	19 (0.8)
No	15,688 (99.6)	13,243 (99.6)	2,445 (99.2)
Bleeding disorder, n (%)				<0.001
Yes	393 (2.5)	283 (2.1)	110 (4.5)
No	15,362 (97.5)	13,008 (97.9)	2,354 (95.5)
Weight loss,‡ n (%)				<0.001
Yes	592 (3.8)	456 (3.4)	136 (5.5)
No	15,163 (96.2)	12,835 (96.6)	2,328 (94.5)
CHF, n (%)				<0.001
Yes	82 (0.5)	58 (0.4)	24 (1.0)
No	15,673 (99.5)	13,233 (99.6)	2,440 (99.0)
Steroid use,§ n (%)				<0.001
Yes	933 (5.9)	703 (5.3)	230 (9.3)
No	14,822 (94.1)	12,588 (94.7)	2,234 (90.7)
COPD, n (%)				<0.001
Yes	1,374 (8.7)	1,090 (8.2)	284 (11.5)
No	14,381 (91.3)	12,201 (91.8)	2,180 (88.5)
Coronary artery disease, n (%)				<0.001
Yes	2,307 (14.6)	1,813 (13.6)	494 (20.0)
No	13,448 (85.4)	11,478 (86.4)	1,970 (80.0)
Diabetes, n (%)				<0.001
Yes	2,715 (17.2)	2,204 (16.6)	511 (20.7)
No	13,040 (82.8)	11,087 (83.4)	1,953 (79.3)
Dialysis, n (%)				<0.001
Yes	66 (0.4)	42 (0.3)	24 (1.0)
No	15,689 (99.6)	13,249 (99.7)	2,440 (99.0)
DVT, n (%)				<0.001
Yes	921 (5.9)	715 (5.4)	206 (8.4)
No	14,822 (94.1)	12,565 (94.6)	2,257 (91.6)
Alcohol use, n (%)				0.934
Yes	520 (3.3)	438 (3.3)	82 (3.3)
No	15,235 (96.7)	12,853 (96.7)	2,382 (96.7)
Hypertension, n (%)				<0.001
Yes	8,605 (54.6)	7,123 (53.6)	1,482 (60.1)
No	7,150 (45.4)	6,168 (46.4)	982 (39.9)
Wound, n (%)				<0.001
Clean or clean/contaminated	13,122 (83.3)	11,159 (84.0)	1,963 (79.7)
Contaminated/dirty/infected	2,633 (16.7)	2,132 (16.0)	501 (20.3)
Peripheral vascular disease, n (%)				<0.001
Yes	386 (2.5)	284 (2.1)	102 (4.1)
No	15,369 (97.5)	13,007 (97.9)	2,362 (95.9)
Pneumonia, n (%)				<0.001
Yes	22 (0.1)	11 (0.1)	11 (0.4)
No	15,733 (99.9)	13,280 (99.9)	2,453 (99.6)
Preoperative transfusion, n (%)				0.002
Yes	109 (0.7)	80 (0.6)	29 (1.2)
No	15,646 (99.3)	13,211 (99.4)	2,435 (98.8)
Sleep apnea, n (%)				0.041
Yes	3,151 (20.0)	2,621 (19.7)	530 (21.5)
No	12,604 (80.0)	10,670 (80.3)	1,934 (78.5)
Functional status, n (%)				<0.001
Dependent	359 (2.3)	242 (1.8)	117 (4.8)
Independent	15,379 (97.6)	13,040 (98.2)	2,339 (95.2)
Approach, n (%)				<0.001
Minimally invasive	10,651 (67.6)	9,308 (70.0)	1,343 (54.5)
Open	5,104 (32.4)	3,983 (30.0)	1,121 (45.5)
Case duration, n (%)				<0.001
<2 h	5,420 (34.4)	4,719 (35.6)	701 (28.5)
2–3 h	5,172 (32.9)	4,389 (33.1)	783 (31.9)
>3 h	5,135 (32.7)	4,163 (31.4)	972 (39.6)

^*Chi-square tests were used to compare categorical variables.

^†Cigarette use is defined as having smoked cigarettes at any point in the year before operation.

^‡Weight loss is defined as 10% of body weight lost in 6 months before operation.

^§Steroid use is defined as use or of oral or parenteral steroids within 30 days before operation.

ASA, American Society of Anesthesiologists; CHF, congestive heart failure; COPD; chronic obstructive pulmonary disorder; DVT, deep venous thrombosis.

Table 2.

Patient Characteristics Associated with Development of Postoperative Complication

Characteristic	aOR (95% CI)	p Value
Age	1.01 (1.00 – 1.01)	0.001
Sex, female	1.16 (1.05 – 1.28)	0.003
Race
White	Ref	—
Black	1.09 (0.93 – 1.28)	0.273
Other	0.95 (0.61 – 1.47)	0.804
Insurance
Commercial	Ref	—
Other	1.30 (1.16 – 1.46)	<0.001
Weight
Underweight	0.95 (0.70 – 1.28)	0.718
Normal weight	Ref	—
Overweight	0.89 (0.79 – 1.01)	0.079
Obese	1.00 (0.93 – 2.62)	0.959
Cigarette use*	1.00 (0.88 – 1.13)	0.959
ASA classification
ASA 1	Ref	—
ASA 2	1.23 (0.73 – 2.05)	0.439
ASA 3	1.56 (0.93 – 2.62)	0.095
ASA 4	2.28 (1.30 – 4.00)	0.004
Ascites	1.25 (0.66 – 2.36)	0.480
Bleeding disorder	1.51 (1.17 – 1.94)	0.002
Weight loss†	1.24 (1.00 – 1.55)	0.056
CHF	1.08 (0.62 – 1.87)	0.794
Steroid use‡	1.63 (1.36 – 1.94)	<0.001
COPD	1.08 (0.92 – 1.26)	0.359
Coronary artery disease	1.18 (1.03 – 1.35)	0.017
Diabetes	1.08 (0.95 – 1.23)	0.214
Dialysis	1.77 (1.00 – 3.15)	0.051
DVT	1.13 (0.95 – 1.36)	0.170
Alcohol use	1.09 (0.83 – 1.41)	0.538
Hypertension	0.98 (0.87 – 1.09)	0.672
Wound
Clean or clean/contaminated	Ref	—
Contaminated/ dirty/infected	1.21 (1.07 – 1.37)	0.003
Peripheral vascular disease	1.33 (1.02 – 1.73)	0.033
Pneumonia	2.72 (0.98 – 7.53)	0.053
Preoperative transfusion	1.24 (0.76 – 2.04)	0.376
Sleep apnea	1.02 (0.90 – 1.15)	0.816
Functional status
Dependent	1.72 (1.34 – 2.23)	<0.001
Independent	Ref	—
Surgical approach
Minimally invasive	Ref	—
Open	1.90 (1.71 – 2.11)	<0.001
Case duration, h
<2	Ref	-
2–3	1.35 (1.19 – 1.53)	<0.001
>3	1.81 (1.60 – 2.06)	<0.001

^*Cigarette use is defined as having smoked cigarettes at any point in the year before operation.

^†Weight loss is defined as 10% of body weight lost in 6 months before operation.

^‡Steroid use is defined as use or of oral or parenteral steroids within 30 days before operation.

aOR, adjusted odds ratio; ASA, American Society of Anesthesiologists; CHF, congestive heart failure; COPD; chronic obstructive pulmonary disorder; DVT, deep venous thrombosis.

Risk- and reliability-adjusted hospital variation

The risk- and reliability-adjusted complication rate by hospital ranged from 8.7% to 30.2%, with a mean of 15.8% (Fig. 1).

Figure 1.

Distribution of risk- and reliability-adjusted hospital complication rate.

Predicted probability of complication

If the least comorbid patient were to undergo colectomy at the highest performing hospital with the highest performing surgeon, the estimated risk of complication would be 1.5% (CI 0.7–3.1%; Fig. 2). If this same patient were to have surgery at the lowest performing hospital, with the lowest performing surgeon, the predicted risk of complication would be 6.6% (CI 3.2–12.2%). By contrast, if the most comorbid patient were to have surgery at the highest performing hospital with the highest performing surgeon, their risk of complication would be 66.3% (CI 39.5–85.6%), compared with a risk of 89.4% (CI 73.7–96.2%) if they were to undergo surgery at the lowest performing hospital by the lowest performing surgeon.

Figure 2.

Predicted probabilities of complication.

Proportion of variance

Fig. 3 shows the VPC of the patient, surgeon, and hospital levels. The proportion of total variance in the development of any non-UTI complication attributable to the patient level was 35.0%, whereas the proportions attributable to the surgeon and hospital levels were 2.4% and 1.8%, respectively (Fig. 3).

Figure 3.

Variance portioning coefficients (VPC) of patient-, surgeon- and hospital-level factors. The VPC ranges from 0 to 1 and is a proportion measuring the level-specific contribution towards the total outcome variance.

Sensitivity analyses evaluating the proportion of total variance in the development of minor and serious complications revealed similarly low contributions from the surgeon and hospital levels. The proportion of total variance in the development of a serious complication attributable to the patient, surgeon, and hospital level was 31.2%, 1.7%, and 2.6%, respectively. For minor complications, the proportion of total variance attributable to the patient, surgeon, and hospital-level was 25.1%, 2.5%, and 1.7%, respectively.

DISCUSSION

This study evaluated the relative contribution of the patient, surgeon, and hospital to postoperative outcomes after elective colectomy. Overall, compared with the surgeon- and hospital-level variance combined, variance at the patient level contributed more than 8-fold more to the development of complications. The impact of patient comorbidities is well illustrated through estimation of complication risk for the least and most comorbid patients in our cohort. The least comorbid patient, regardless of surgeon or hospital, had a risk of developing a complication after elective colectomy of less than 10%. However, for the most comorbid patient, the risk of complication was greater than 60%, whether they are treated at the highest performing hospital with the highest performing surgeon or the lowest. Our results underscore the need for patient-level interventions targeting modifiable risk factors within our statewide health system.

Examples of such initiatives include mandatory preoperative smoking cessation, as well as “prehabilitation” programs, which may improve nutritional status and exercise capacity before an operation.^19–21 Although the data for prehabilitation are not conclusive, some studies have shown significant improvement in clinical outcomes across a wide range of operations and have the added benefit of lower episode costs.^19,20 Further, prehabilitation programs have been shown to be affordable, costing roughly $100 per patient, which is notably lower than the $150 price tag of enhanced recovery protocols.^19,22 Efforts aimed at widespread implementation of patient-centered prehabilitation programs could have significant impact.

Although the depth and intensity of preoperative risk modification will vary based on operative indication and timeline, addressing a patient’s foundational health no matter the length of the preoperative period is paramount. Not leveraging the surgical episode to help patients improve their foundational health is a missed opportunity. For example, the benefits of smoking cessation can be appreciated in as little time as 12 hours, and the cardiopulmonary benefits of prehabilitation have been seen in patients with as little as 1 week prior to operation.^23,24 Patients who have more time prior to operation may benefit from longer prehabilitation, but this must be weighed against the risk of decreased patient compliance, because many patients do not want to delay surgery. Further, there may exist ways to efficiently optimize patients that remain undefined. This area of research should be prioritized given the substantial contribution of patient factors towards the variation in outcomes. At our institution, we aim for a prehab period of 4 weeks, and through our program we have been able to achieve improved physiologic tolerance of surgery and lower postoperative complication rates in those undergoing major abdominal surgery.¹⁹

Beyond prehabilitation, to truly move the needle on reducing poor outcomes after surgery, healthcare systems must make a comprehensive commitment to address a patient’s foundational health at every interaction. This requires united efforts and messaging from all providers, including surgeons. In particular, modifying poor health behaviors and addressing social needs will be critical. These factors are the most impactful drivers of health outcomes and therefore merit early identification and intervention. Efforts in these areas are likely to both improve surgical outcomes and preexisting comorbidities, yet previous work has shown that there is significant room for improvement. For example, Delaney and colleagues²⁵ showed that surgeon adherence to optimization practices (weight loss, smoking cessation) before elective hernia repair is only 51% to 76%. Although a lofty goal, future surgeon- and hospital-level quality improvement efforts should target these practices to ensure operative risk is minimized as much as possible, especially for nonurgent colon surgery.

The surgical episode may actually be a profoundly meaningful time to motivate behavior change and modify social determinants of health (SDOH). At our institution, we are piloting a novel surgical pathway²⁶ that leverages surgery as a teachable moment during which patients have an increased likelihood of achieving sustained health behavior change.²⁷ For instance, one study showed that undergoing an operation independently increases the likelihood that a patient quits smoking.²⁸ Our initiative addresses patients’ foundational health by screening them for poor health behaviors and social determinants of health during their surgical episode and connecting them to meaningful resources. By integrating intervention this into the surgical pathway, we believe patients can have improved surgical outcomes, less burdensome comorbidities, and better health behaviors.

Our findings do not suggest that surgeon and hospital factors are irrelevant to patient outcomes; rather, the variation in quality of care administered across surgeons and hospitals does not explain the majority of variation seen in complications rates. Therefore, this is not a criticism of past quality improvement efforts; instead, this study may provide evidence that previous initiatives have been successful in optimizing and standardizing the procedural elements of care delivered by surgeons and within hospital systems. In this context, these data are consistent with previous work showing that the surgeon contribution to the development of complications is relatively low, ranging from 0.7% for readmission to 4.5% for surgical site infection.¹¹

Within our study, nearly 60% of risk was unaccounted for after considering patient comorbidity, surgeon performance, and hospital setting. It is likely that at least a portion of this risk can be attributed to SDOH that were unmeasured in our model. Although the impact on overall health has been understood for more than a decade, with some estimates that SDOH drive up to 40% of health outcomes,²⁹ more recent work has begun to focus on the specific impacts on surgical patients. For example, food insecurity, which impacts 10% to 15% of households in the US,³⁰ has been associated with longer lengths of stay in patients requiring acute care hospitalizations and sameday surgeries.³¹

There are several limitations to this study. First, this analysis included patients undergoing elective colectomies only, so there is limited understanding of relevance to colectomies performed in an emergent setting or to other operations altogether. Second, our results may not be generalizable to other procedures, and the relative contribution of the patient, surgeon, and hospital towards the development of postoperative complications may be different. However, the contribution of surgeon- and hospital-level factors has previously been shown to be low across a broad range of operations.¹¹ This suggests targeted quality initiative interventions to modify patient risk may be worthwhile before most, if not all, procedures. Third, the database used comprises a sampling of cases and therefore may not completely represent each individual surgeon’s actual case mix. Finally, this study used data from a regional, state-wide quality collaborative, participation in which has been associated with improvements in clinical outcomes,³² which may limit the generalizability of this data to hospital systems outside of the state.

CONCLUSIONS

In conclusion, our results suggest that patient-level factors drive the development of postoperative complications substantially more than surgeons or hospitals. These data suggest that future initiatives aimed at improving surgical care should focus on optimizing patient health through prehabilitation, modification of poor health behaviors, and addressing social determinants of health.

Abbreviations and Acronyms

aOR

adjusted odds ratio

CPT

current procedural terminology

CQI

Collaborative Quality Initiative

ERP

enhanced recovery protocol

MI

myocardial infarction

MSQC

Michigan Surgical Quality Collaborative

SDOH

social determinants of health

SSI

surgical site infection

UTI

urinary tract infection

VPC

variance partitioning coefficient