Introduction

Over the last 25 years, the US prevalence of obesity has increased dramatically in all age groups (CDC). More than 400 million adults worldwide are considered obese, with body mass indices (BMI) >30, and nearly 1.6 billion are overweight, with BMI’s between 25 and 30.1 Despite its ubiquity, obesity has been linked to increased risk for developing serious pathologies, including breast, prostate colorectal, endometrial, and other cancers.2–4 It is simultaneously a common risk factor for chronic cardiovascular disease and diabetes.5–7 Treatment for obesity-associated illnesses consumes 10% of all US healthcare expenditures,8 and increased BMI has been linked to poorer outcomes across the board.

More specifically, obesity has been shown to predict the incidence of and poor outcomes from pancreatic cancer,9–12 as well as to confer a higher risk of developing chronic pancreatitis.13 In both cases, surgical pancreatic resection is imperative. However, obesity has also been linked to a variety of peri- and post-surgical complications following pancreatectomy, including increased length of hospital stay (LOS),14–17 surgical site infection,18–20 blood loss,21,22 and decreased disease-free and overall survival.15 Conversely, though increased BMI is frequently cited as a risk factor for deleterious peri-surgical outcomes in patients undergoing pancreatic resection, its application as a universal predictor seems to be unfounded. While several studies have directly examined the impact of BMI on outcomes following pancreatic resection,14–17,21,23,24 their small sample sizes and inconsistent findings have limited the power of their conclusions. Additionally, their results are frequently contradicted within the literature, and their analyses are often limited by a lack of specificity toward BMI-associated outcomes.

However, recent increases in resistant gram-positive and gram-negative bacterial infection and colonization both nationally and within hospitals suggest that identifying predictors of complications before surgery may limit the risk of hospital-acquired morbidity and excess mortality.25–28 Because 5-year survival following pancreatic resection stands at less than 20%,29 with post-surgical morbidity as high as 50%,14 an accurate assessment of BMI-associated complications in pancreatic resection may facilitate targeted pre- and post-operative therapy, greatly improving outcomes and reducing costs in this sick cohort. We undertook this study to identify BMI-associated outcomes in patients undergoing pancreatic resection.

Materials and Methods

This review and meta-analysis was performed using the PRISMA criteria to ensure validity and transparency. We conducted a MEDLINE search using the Boolean operator “AND” to join all combinations of “BMI” or “obesity,” with “pancreatectomy,” “pancreatoduodenectomy,” or “pancreaticoduodenectomy.” We did not restrict our search by study design or to core clinical journals, only restricting our initial inquiry to publications in English between January 1990 and June 2010. We included in our review all studies providing information on BMI-associated outcomes in human patients undergoing pancreatic resection for all indications, defining pancreatectomy as an intervention involving any form of pancreatic resection. We assumed that outcomes reported in primary studies were based on individual patients undergoing a single pancreatic resection procedure.

We excluded a priori all case reports/series, reviews not providing new information, studies not assessing outcomes for pancreatic resection, studies failing to provide information about pre-surgical BMI, and studies with equivalent experimental and control group BMIs. A priori sensitivity analyses were not specified, and we did not generate explicit criteria for identifying and excluding primary studies with inherent biases. Our sensitive, rather than specific, inclusion criteria were intended to generate a large study cohort with which to assess a variety of BMI-associated outcomes. Both investigators assessed all articles for failure to meet these criteria. Disagreement was resolved by reaching consensus. All outcomes reported in primary studies are reported with their original qualifying data.

From the studies meeting the inclusion criteria (n = 17), both reviewers independently extracted data regarding indication for surgery, type of intervention, patient BMI, all surgery-associated outcomes, and any other pertinent background information. For descriptive BMI-associated outcome analyses, we separated patients into BMI-unit intervals for normal weight (BMI 18–25), overweight (25–30), and obese (>30) patients. Because BMI was reported inconsistently across the primary study cohort, our analyses frequently evaluated BMI <25 versus >25, and BMI >30 versus <30, though we compared all three tiers whenever possible. Several studies reporting BMI as a continuous variable are included in qualitative analyses.

For our meta-analysis, we created pooled rates for all binary outcomes where at least three studies reported those outcomes with corresponding BMI information. We performed Fisher exact tests using R (version 2.12) to identify any differences in outcomes across BMI tiers. We reported the results of these analyses as odds ratios (OR) with 95% confidence intervals and utilized the p < 0.05 level to specify statistical significance. Similarly, we performed a subgroup analysis of studies using the International Study Group on Pancreatic Fistula (ISGPF) definition of pancreatic anastomosis to illuminate any disparities in the measurement of this outcome among studies not reporting utilization of this standard. All other outcome analyses used definitions of those outcomes used within each individual study, and there was insufficient data to perform sensitivity analyses on these outcomes.

For continuous variables, because we could not extrapolate reported averages backward to their corresponding patients, we created weighted totals by multiplying each reported estimate by its primary-study sample size and dividing that by the pooled sample size. This metric emphasized the importance of sample size in our aggregate estimate for each continuous variable. Since continuous outcomes in primary studies were peri-surgical estimates only, our metric and analyses maintained the descriptive nature of these outcomes. Consequently, we performed minimal statistical analyses of continuous variables. Outcomes that were quantified in only one or two studies are reported separately. Qualitative analyses include the results from univariate and multivariate analyses of BMI-associated surgical outcomes within each primary study.

Results

Study Population

Of 71 studies identified in our broad initial search, we excluded 53 from our review and meta-analysis: those that did not include data from human patients (n = 11),30–40 case reports/series (n = 15),41–55 review articles not providing new information (n = 7),56–62 those not assessing outcomes for pancreatic resection (n = 11),63–73 those failing to provide information about pre-surgical BMI (n = 7),74–80 and those with equivalent experimental and control group BMI’s preventing the extraction of outcomes related to different BMIs (n = 2).81,82

In all, 17 studies were included in the final set (Table 1).14–17,21–24,83–91 All primary studies reported retrospective analyses of prospectively maintained databases, suggesting that selection bias may exist in this body of literature. Our 17 study cohort included a total of 4,045 patients, though BMI information was reported heterogeneously. Four studies split BMI into three tiers and yielded 881 patients with a stratified BMI <25, 592 patients with a BMI >30 and 546 patients with a BMI between 25 and 30.16,21,22,24 Eleven other studies assessed BMI only as >30 versus <30 or >25 versus <25, and these studies provided 1,309 patients with BMI <30 and 328 patients with BMI >25.14,15,17,23,83–91 One study assessed outcomes for patients with BMI >27 versus <27, and 216 patients therein with BMI >27 were included in our >25 comparator group, though we could not include the remaining 213 patients with BMI <27 from the primary study in any of our analyses.88 In aggregate, these data yielded a sum total of 2,736 patients with BMI <30, 1,682 with BMI >25, and 546 with BMI definitively between 25 and 30 for various quantitative analyses. Two other studies evaluated BMI as a continuous function and data therein were included only in our qualitative analysis.84,87 One study provided a stratified BMI measurement, but we could not extract outcomes data corresponding to patients in each BMI tier, and we included analysis offered by this study in our descriptive analysis only.90 There was insufficient information to generate cohorts of underweight (BMI <18) or class II and III obese patients.

Table 1 Characteristics of 17 studies evaluating BMI-associated morbidity in pancreatectomy
Full size table

Surgical indications and interventions were diverse across primary studies. This cohort included 2,753 patients with adenocarcinoma, 216 with neuroendocrine neoplasms, 339 with pancreatitis (chronic and immune), 93 with benign neoplasms, and 965 with other operative indications (Table 2). Some patients may have presented with more than one indication for pancreatic resection, but we could not identify outcomes for these patients individually. Reported surgical interventions included classic PD (n = 1,001), pyloris-preserving PD (PPPD; n = 1,365), unspecified PD (n = 1,047), unspecified proximal pancreatectomy (n = 100), unspecified distal pancreatectomy (n = 756), unspecified partial pancreatectomy (n = 17), and total pancreatectomy (n = 17). This diversity of indication and intervention was present within many individual studies.16,20,22,23,64,68,73,75,77,79,80,82,83,85–90 Consequently, we could not stratify outcomes by these conditions, though more research is necessary to illuminate the possibility of disparate outcomes associated with high BMI in patients undergoing different types of pancreatic resection and with different underlying pathologies.

Table 2 Pooled indications for surgical intervention
Full size table

Outcomes

Eleven studies included BMI as a covariate in multivariate analyses (Table 3),15,16,21,23,24,83,85,86,88–90 but only PF emerged consistently as a clinically relevant complication of high BMI. Four studies found association between increasing BMI and PF in multivariate analyses, two with BMI >30, one with BMI >27, and one with BMI >25 (range OR 1.6–4.2).23,83,88,89 Conversely, one study found no association between BMI >25 and PF and another between BMI >30, 25–30, or <25 and PF.24,85 Of five studies utilizing the ISGPF definition of PF,23,83,85,86,89 three found a multivariate association between PF and BMI, one found no association, and one did not include a multivariate analysis. Two studies found an association between clinically relevant PF and BMI >27 (OR 3.4) and >30 (OR 6.5).83,88

Table 3 Morbidity predicted by BMI in multivariate analyses
Full size table

Pooled analyses of PF by pre-surgical BMI, including 485 patients from 11 studies, found significant association between increasing BMI and PF (Table 4).16,17,21–24,83,85,86,88,89 Overall PF rates ranged from 4.7% to 31.0%. Eighty-five patients in seven studies had BMI <25, while 105 patients in those studies had BMI >25.16,21–24,85,86,91 This corresponded to PF rates of 8.6% and 12.2% (p = 0.01), respectively, both of which are slightly lower than the 25% PF rate established by Pratt and colleagues in 233 consecutive PDs.92 In six studies evaluating BMI >30, 106 patients had BMI >30 and 283 patients had BMI <30, with PF rates of 22.9% and 15.8% (p < 0.001), respectively.16,17,21,24,83,89 Three studies utilizing the ISGPF definition of PF had pooled rates of 12.6% and 30.0% for BMI <25 and >25 (p < 0.001).23,85,86 Two studies using the ISGPF definition of PF for BMI >30 and <30 had pooled rates of 36.9% and 21.1% (p < 0.001), respectively.83,89 However, none provided information about differential outcomes for PF in high-BMI versus low-BMI patients.

Table 4 Pooled analyses of BMI-associated morbidity and surgical site infection
Full size table

Other BMI-associated outcomes were less consistent. One study found a multivariate association between BMI >27 and both any (OR 1.8) and major (OR 3.3) complications.88 Two studies contradicted this, one finding no association between increasing BMI and any complication (OR 1.0) and the other finding no association between increasing BMI and major complications.16,24 However, the definition of complication was rarely reported and probably varied across primary studies.

Ultimately, seven studies reported any complication as an outcome of high BMI with rates from 11.6% to 54.3%.14–17,21,24 Three studies assessing BMI <25 for any complication had pooled rates for BMI <25 and >25 of 40.6% and 39.5%, respectively (p = 0.69).16,21,24 Seven studies assessing BMI >30 by any complication had pooled rates for BMI >30 and BMI <30 of 43.7% and 35.5% (p = 0.002).14–17,21,24,88 One study reported a significantly decreased multivariate risk of disease-specific mortality with increasing BMI (OR 0.74), but this corresponded to a significant univariate risk increase for any complication with increasing BMI.21 With complication inconsistently defined, no distinct pattern was visible in these data.

Some less subjective outcomes found agreement across the primary study cohort, but the clinical relevance of any differences seen was questionable. A single study evaluated multivariate association between surgical site infection and BMI.24 While there was slight association between BMI >30 and increased risk of infection over BMI <25 (OR 1.10, p = 0.03), there was no association between BMI 30–25 and increased infection rates. Our pooled analysis yielded a similar result (22.9% and 15.8% for BMI >30 and <30 (p < 0.001), 8.0% and 9.3% for BMI <25 and >25 (p = 0.40)), with infection rates ranging from 4.6% to 14.0% in six studies reporting that outcome.14,16,17,21,24,86 Two studies reporting serious infection as an outcome (bacteremia, pneumonia, intra-abdominal infection) found rates of 26.3% versus 51.9% (p = 0.01) in patients with BMI <25 and >25, and one study found rates of 15.1% versus 15.8% (p = 1.00) in patients with BMI <30 and >30.17,22,86 However, Williams and colleagues attribute their equivalent rates of post-operative infections across BMI categories to their meticulous use of appropriate prophylactic antibiotics.16 A prophylactic regimen not specifically targeted to patients with more widely distributed adipose tissue could reduce serum drug concentrations and, consequently, therapeutic efficacy. Differential prophylactic utilization across primary studies may have contributed to differences seen in infection rates across BMI categories.

Similarly, while Fleming and colleagues report multivariate association between both positive lymph node status (OR 2.45, p = 0.02) and cancer recurrence (OR 2.00, p = 0.05) in patients with BMI >30, they note that decreased use of pre-operative therapy in patients with BMI >35 in their study population significantly increased the likelihood of these outcomes (p = 0.02).15 Additionally, Su and colleagues note that the bacteria colonizing patients at time of surgery rarely corresponded to bacteria causing post-operative infection, suggesting that prophylactic antibiotics impacted outcomes in unpredictable ways.22 Though these factors were unaccounted for in nearly all studies assessing BMI’s impact on the incidence of both surgical site infection and tumor size, inconsistent or withheld pre-operative therapy may explain increased risk for these outcomes among obese and overweight patients.

Qualitative analyses were similarly inconclusive. Six studies evaluated estimated peri-surgical blood loss as an outcome, with medians from 400 to 1,000 ml (Table 5).14–16,21,22,24 Weighted overall blood loss estimates (661.2 versus 801.3 ml for BMI <25 and >25 and 615.9 versus 790.1 ml for BMI <30 and >30) corresponded well with each other and reflected consistent increases with increasing BMI seen in all studies reporting this outcome. Similarly, in four studies reporting median operative time (range 363 to 439 min), all found increased operative time with increasing BMI.14–16,21 Weighted operative times (363.8 versus 382.7 min for BMI <25 and >25 and 351.6 versus 368.7 min for BMI <30 and >30) were similarly disparate, though the small magnitude increases likely reflect increased care by the surgeon and may have little clinical relevance.

Table 5 Weighted estimates for continuous variable outcomes
Full size table

No clear relationship emerged between BMI and LOS. Five studies reported this outcome, medians ranged from 8 to 11 days, and means were predictably higher.14–17,21 All median LOS were similar to those noted by Cameron and colleagues.29 Weighted LOS for BMI <30 and >30 were 10.0 versus 9.8 days, respectively. However, disparities were within 1 day for all BMI categories in three of four studies, with Williams and colleagues noting a 1.5-day difference between BMI >30 and the other categories.16 However, Williams and colleagues found 2× increased risk for delayed gastric emptying (DGE) among patients undergoing classic PD (3% versus 7%). Additionally, they note a procedural transition from classic PD to PPPD, a substantial increase in surgical volume during the study period, and a significantly increased use of biliary stents in patients with BMI >30 (p = 0.001).16 Disparate resection procedure utilization and temporal imbalances in BMI distribution could explain the small difference seen in pooled LOS.

With conflicting results, two other studies reported DGE as an outcome (42% versus 53% for BMI <25 and >25, 18% versus 11% for BMI <30 and >30).17,86 Both studies had small sample sizes (n = 92 and n = 85) and neither reported univariate significance in this association. Similarly, two studies offered multivariate analyses concluding no association between BMI and DGE, though BMI was not quantified with number of patients experiencing this outcome in one of the two.83,86 Two studies reported decreased glucose control following pancreatectomy, one noting a significant increase in diabetes among patients with higher BMI (p = 0.03) and the other a loss in glucose control (r = 0.41, p = 0.01).84,87 However, average BMI for developing diabetes versus not developing diabetes was 24.1 versus 21.9; as the majority of patients in each category are within normal BMI range, the clinical importance of this finding is questionable. Additionally, as neither study included a multivariate analysis of these outcomes, confounding of the data is likely.

Soft pancreatic consistency is widely touted as predictive of PF following pancreatectomy, and five studies addressed this outcome in relation to BMI, though no surgically useful patterns emerged.17,21,23,24,85 Gaujoux and colleagues found that BMI >25 was predictive of both fatty pancreas and absence of pancreatic fibrosis.23 These factors were all individually predictive of PF. Age was an independent predictor of fatty infiltration (p = 0.03), but BMI was significantly associated with soft pancreatic remnants (p = 0.04). However, soft consistency most highly correlated with the absence of pancreatic fibrosis (p < 0.0001), which Rosso and colleagues confirm (p < 0.0001).85 Three other studies confirm that increasing BMI is associated with increased fat infiltration, two of the pancreas and one in the retroperitoneal space.17,24,85 However, Rosso and colleagues note that soft pancreatic parenchyma was not associated with increased fatty infiltration (p = 0.17). Conversely, Tsai and colleagues report that BMI was not associated with soft pancreatic tissue (p = 0.23), though they did not evaluate the study population’s level of parenchymal fibrosis.21

Comprehensive survival analysis was impossible due to incomplete reporting, though no increased risk with high BMI was evident. As noted above, a single study found multivariate association between decreased risk of disease-specific mortality and increasing BMI (OR 0.74, p < 0.01), as well as a significant increase in 5-year survival in patients with BMI >30 and 30–25 (22% and 22% versus 15%, p = 0.02).21 Fleming and colleagues did not find significant association between BMI >35 and decreased survival in a multivariate analysis (p = 0.08).15 Similarly, Benns and colleagues found no significant association between either disease-free or overall survival in patients with BMI ><30 (p = 0.50 and p = 0.46, respectively).14 Tsai and colleagues noted that survival was similar among 32 underweight patients and the remaining normal patient cohort (n = 398), though this was the only evaluation of outcomes in underweight patients across the study cohort.21

Peri-operative mortality was extremely low (range 0.0% to 4.4%) in the four studies reporting it, and none found association between increasing BMI and the incidence of mortality.17,21,88,91 In multivariate analysis, Bentrem and colleagues found that patients undergoing PD with BMI >30 were more likely to be admitted to the ICU (p = 0.003), that BMI >30 was associated with delayed ICU admission from the surgical ward, and that ICU admission was associated with overall decreased survival (p < 0.0001), but they did not identify an association between BMI and decreased survival.90

Discussion

We have shown that most differential outcomes between high-BMI and low-BMI cohorts undergoing pancreatic resection are far lower in magnitude and far better in prognosis than might be expected. Though several morbidity indices were slightly higher in overweight and obese patients, they rarely reached clinical significance. Higher rates overall of pancreatic fistula in patients with higher BMI are worrisome, but several studies finding no association between BMI and pancreatic fistula rates may offer valuable insight into best-practices scenarios. Consistent use of prophylactic therapeutics, increased peri-surgical care, and better post-surgical management in obese and morbidly obese patients maybe valuable tools for reducing small, but clinically significant disparities in outcomes between patients with different BMIs in this cohort.

Unlike previous studies in this genre, by pooling BMI categories from 17 studies, we were able to establish a large enough cohort to perform a stringent and satisfactory statistical review and meta-analysis for a variety of endpoints previously found to be associated with high BMI. Though outcomes including incidence of PF, surgical site infection, blood loss, and operative time emerged as consequences of increasing BMI in patients undergoing surgical pancreatic resection, outcomes including post-operative LOS, tumor size, harvested lymph nodes, delayed gastric emptying, peri-operative mortality, and decreased overall survival exhibited irregular association or no association with high BMI. These clinically relevant conclusions suggest that BMI alone should not preclude surgical pancreatic intervention.

Importantly, we elucidated the fact that obesity is not a universal predictor of poor outcomes in surgical patients. By rigorously excluding specious outcomes associated with historically significant risk factors like high BMI, therapy may be precisely targeted to individual patients. Because meticulous consideration of all possible risk factors by physicians is a noted contributor to the perfect storm of healthcare over-utilization in the USA,93 identifying high BMI as a null predictor in patients undergoing pancreatic resection may greatly reduce systemic costs. More importantly, identifying a lack of association between high BMI and many poor surgical outcomes may illuminate imperative post-operative therapeutic considerations, systemically resulting in improved surgical outcomes.

Our study has several limitations grounded within primary study designs and reporting variations. Most significantly, we cannot exclude the possibility of selection bias from individual studies. Each of the 17 studies included in our analysis was a retrospective assessment of prospectively collected data. No randomization was used, and all patients included in our final analysis were selected by individual surgeons to undergo pancreatic resection. However, as we did not exclude studies based on study design, this represents the entire body of literature on this topic. Though targeted pre-operative prophylaxis, peri- and post-operative care may improve outcomes in patients with high BMI, there may be inherent physiological differences, including increased peri-pancreatic fat leading to higher rates of PF that could not be elucidated due to biased inclusion criteria within primary studies. Unfortunately, the direction this bias may take is not predictable from these data.

Additionally, there was no consistency across indicators for surgical intervention within the primary cohort, nor were outcomes reported universally or homogenously. Comorbidities, including diabetes rates, were similarly inconsistently reported, suggesting the possibility of confounding. However, diverse surgical indications and comorbidities are commonplace in this body of literature. To mitigate potential disparities in patient characteristics within individual primary studies, we restricted our pooled analyses to outcomes reported in at least three studies. Additionally, because we could not illuminate the internal architecture of reported continuous variables, we limited statistical testing to variables with discrete expression and conclude significance only with 95% confidence. Unfortunately, incomplete reporting prevented us from identifying any differences in outcomes associated with high BMI in different interventions, but we do not suspect that BMI would substantially increase a patient’s risk for complications in one high-risk surgical pancreatic intervention over another.

To mitigate the possibility of serious consequences associated with varied reporting across primary studies because PF is one of the most common and consequential outcomes of pancreatectomy in all patients,92 we performed a sensitivity analysis using the cohort of studies adopting the ISGPF definition of PF. We found results similar in direction and magnitude to those of our complete study population, suggesting some cohesion among primary study outcome measurements despite potential measurement inconsistencies. Finally, though our study population included more than 4,000 patients and at least 592 with a BMI >30, limited reporting prevented us from creating a truly sound metric for predicting outcomes associated with specific BMI levels. Further research is needed to ensure therapy meets the prophylactic needs of patients, including underweight and class II and III obese surgical patients who we were unable to assess in this review and meta-analysis.

Conclusion

This review identified a clear association between BMI and PF incidence. Given the consistent lack of PF definitions and BMI gradations in these studies, the clinical severity of high-BMI-associated PF could not be ascertained reliably, however. BMI was not found to be associated with LOS, hospital mortality, disease-free survival, or overall survival in a combined analysis of the studies recording these outcomes. Thus, though it appears BMI increases the operative complexity of pancreatic resection, most associated increases in peri- and post-operative morbidity can potentially be mitigated with surgical care and an aggressive patient management schedule. Further research is necessary to explore the impact of BMI on specific pancreatic surgical interventions, as well as the impact of BMI on long-term outcomes following pancreatic resection.