The Effect of Laparoscopy on Survival in Pancreatic Cancer FREE

THE POPULARITY of laparoscopic surgery has grown substantially during the past decade.¹ Although some benefits of laparoscopic surgery, such as decreased operative morbidity and shorter recovery, have been increasingly recognized, many observers have become concerned about the use of laparoscopic surgery in the management of gastrointestinal cancer.^2- 3 Considerable alarm was generated by case reports of port-site recurrences following laparoscopic cholecystectomy for gallbladder cancer⁴ and of unexpectedly poor outcomes following laparoscopic resection of early-stage colorectal cancer.⁵ Laboratory studies have suggested that carbon dioxide pneumoperitoneum, the most common means of facilitating laparoscopic surgery, may actually promote the growth and dissemination of intraabdominal tumors.⁶

The oncologic efficacy of laparoscopic resection for colorectal cancer is currently being addressed by randomized clinical trials comparing laparoscopic with open surgery.^7- 9 In the meantime, many surgeons are observing a voluntary moratorium on laparoscopic resection for gastrointestinal cancer outside clinical trials. Although reliable data on the safety of laparoscopic cancer surgery in humans are lacking, some insight into the effect of laparoscopic surgery on cancer biology may be gained by evaluating patients with pancreatic cancer. Some patients with pancreatic cancer are exposed to laparoscopy (and therefore carbon dioxide pneumoperitoneum) for either diagnosis or staging at some point during their illness. We used a linked cancer registry and administrative claims database to study patterns of mortality in persons with pancreatic cancer who had either a laparoscopic procedure or a laparotomy but no pancreatic resection, to identify whether laparoscopy had an effect on survival.

We submitted a proposal to use a linked cancer registry and Medicare claims database to test the following hypothesis: in patients with metastatic pancreatic cancer who had either a diagnostic laparoscopy or diagnostic laparotomy but no other procedure (including pancreatic resection or biliary or gastric bypass), exposure to laparoscopic surgery has no effect on survival. Our aim in limiting the sample to individuals with metastatic disease at diagnosis who had no surgery beyond a diagnostic abdominal procedure was 2-fold: (1) to create a cohort that was relatively uniform with respect to important determinants of prognosis; and (2) to minimize confounding due to other interventions that might affect survival. Once we received the files and conducted a preliminary review of the data, it became clear that the number of individuals in the files who satisfied the eligibility criteria was too small to estimate relative survival differences with adequate precision. To improve statistical power, we broadened the eligibility criteria to include patients at any stage at diagnosis and allowed additional palliative surgical interventions, such as biliary and gastric bypass. Because the decision to include these subgroups was made post hoc, we present the results of several different analyses here. In addition to evaluating the entire cohort, we also performed stratified analyses according to the presence of distant metastases at diagnosis and whether additional surgical procedures were done.

The Surveillance, Epidemiology, and End Results (SEER) program actively acquires data from several population-based cancer registries in various geographical regions of the United States that comprise 13.9% of the total population. Detailed cancer data, including tumor sites, stages, histologic findings, and treatment are abstracted from several sources for up to 10 incident tumors per person residing in a SEER area. Vital status is updated annually from a public use file from the National Center for Health Statistics.¹⁰ The SEER data we used for this study were contained in a Patient Entitlement and Diagnosis Summary File (PEDSF), which contains 1 record per individual in the SEER database that was matched with Medicare records.

The Medicare program provides health insurance benefits to 97% of persons 65 years and older, individuals younger than 65 years who are entitled to disability benefits, and most individuals with end-stage renal disease in the United States. To ascertain exposures to surgical procedures and comorbid medical conditions, we used the Medicare Provider Analysis and Review (MEDPAR) file and the outpatient Standard Analytic File. The MEDPAR files contain 1 record per inpatient hospitalization in all short-stay and long-stay hospitals as well as skilled nursing facilities. Each MEDPAR record lists up to 10 surgical procedures and up to 10 medical diagnoses (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] procedure and diagnosis codes). The outpatient Standard Analytic File contains 1 record for each outpatient service claimed to Medicare. Procedures are coded in the Standard Analytic File as Current Procedural Terminology or Health Care Financing Administration Common Procedure Coding System codes.

To facilitate cancer-related health services research, the National Cancer Institute, SEER registries, and the Health Care Financing Administration have collaborated in creating and making available to researchers, a database containing records of individuals contained in the SEER registries who also appear on Medicare enrollment records. Medicare claims have been successfully linked to 93.8% of SEER records.¹¹ Records appearing in SEER and Medicare files are linked by a unique identification number. These linked files permit longitudinal follow up of individuals using Medicare claims data, combined with detailed, population-based oncologic information from the PEDSF file. The analysis and interpretation of these data are the sole responsibility of the authors and do not represent the views of either the National Cancer Institute or the Health Care Financing Administration.

We assembled a cohort of persons aged 65 years and older who were newly diagnosed with pancreatic cancer (International Classification of Diseases for Oncology, Second Edition site codes C25.0-C25.3, C25.7-C25.9), resided in a SEER area, and were captured in both the SEER cancer registry and the Medicare claims database between January 1, 1991, and December 31, 1996. Of the 13 263 persons with pancreatic cancer in the SEER database during this time, 72.9% were 65 years and older. Individuals were eligible for inclusion if they had a diagnostic laparoscopic procedure or a diagnostic laparotomy. Persons who were entitled to coverage by Medicare because of end-stage renal disease or a disability were excluded (3.4% of PEDSF records). Individuals were also excluded if they had a prior malignancy in their PEDSF record (12.2%), missing or incomplete data on tumor extent (32.2%), or a histologic finding that was not adenocarcinoma or a variant (2.4%). Of individuals who had a laparoscopic procedure or a laparotomy, 19.5% were excluded because they had a pancreatic resection or ablative procedure and 34.4% were excluded because they had a cholecystectomy between the years 1991 and 1996.

We considered an individual to be exposed to a laparoscopic surgical procedure if the ICD-9-CM procedure code 54.21 (laparoscopy) or 54.51 (laparoscopic lysis of adhesions) appeared in any MEDPAR record (the Standard Analytic File did not contain any records for these procedures as ambulatory cases). We considered exposure to open surgery to have occurred if the ICD-9-CM codes 52.12 `(open biopsy of pancreas), 54.11 (exploratory laparotomy), or 54.5 (lysis of peritoneal adhesions) were present as a separate procedure. Patients who had both laparoscopic surgery and open abdominal surgery were included in the laparoscopic surgery group since we were primarily concerned about detecting an adverse effect of laparoscopy in this study. Individuals who had a biliary or gastric bypass were included if they also had a diagnostic laparoscopy or laparotomy as a separate procedure.

We used a code-based modification of the Charlson comorbidity index to model comorbid medical conditions,^12- 13 which has been used to adjust for comorbidity in other studies using SEER-Medicare data.^14- 15 The index was further modified by excluding the categories for any malignancy and metastatic tumor. To assess comorbidity, we analyzed ICD-9-CM diagnosis codes appearing on MEDPAR records for hospitalizations that occurred at least 1 year prior to the diagnosis of pancreatic cancer. We categorized comorbidity according to whether individuals had any hospitalizations and calculated aggregate comorbidity scores (using all ICD-9-CM diagnosis codes appearing in any eligible hospitalization) for those patients who had 1 or more prior hospitalizations. In calculating comorbidity scores and in analyses that adjusted for comorbidity, we limited the sample to those persons 66 years and older who were not enrolled in a health maintenance organization (HMO) at the time of diagnosis of their pancreatic cancer (14.9% of individuals in the PEDSF file were HMO enrollees when their cancer was discovered). Since Medicare may not collect a bill for each procedure for individuals enrolled in risk-based HMOs, only patients who had indemnity coverage and were "at risk" of being captured in MEDPAR records could contribute data toward calculating a valid comorbidity score.

Means of continuous variables were compared using the t test and proportions of categorical variables were analyzed using the χ² test or the Fischer exact test. We calculated survival time for each subject from the date of diagnosis of pancreatic cancer until the date of death for patients who died or until December 31, 1996, for patients who had not died by this date. Because the SEER database records only the month and year of diagnosis, we assumed that each case of cancer was diagnosed on the first day of the month of diagnosis in determining survival time. Cumulative mortality was estimated using Kaplan-Meier survival curves¹⁶ and differences between curves were assessed using the log-rank test.¹⁷ The Greenwood formula was used to estimate the SE of the Kaplan-Meier estimate of the survival function.

In testing univariate associations and in building regression models, age was treated as a continuous variable and all other exposures were treated as categorical, with indicator variables used to represent categories as appropriate. Since person-level data on socioeconomic factors are not present in SEER-Medicare files, socioeconomic status was represented in regression models by the median annual household income in a patient's census tract of residence, categorized according to the quartile distribution of the entire cohort. The effects of different factors on survival were estimated using Cox proportional-hazards regression.¹⁸ The relative hazard of death associated with each exposure was expressed as the hazard ratio and 95% confidence interval (CI). Estimates of hazard ratios greater than 1.0 represent an increased rate of death and estimates of less than 1.0 represent a decrease in the rate of death. We evaluated the consistency of the association between laparoscopic surgery and survival by comparing hazard ratios (for laparoscopic surgery compared with no laparoscopic surgery) within selected subgroups. Adjusted risks were estimated using multivariate models. We explored the effect of potentially confounding exposures by fitting models that included relevant sets of predictor variables. All reported P values are 2 sided.

Nine hundred three persons newly diagnosed with primary pancreatic cancer between January 1, 1991, and December 31, 1996, in the SEER areas satisfied our eligibility and exclusion criteria. One hundred twelve persons had laparoscopic surgery recorded as a separate procedure, 27 (24.1%) of whom also had a diagnostic or staging laparotomy. Seven hundred ninety-one had a laparotomy but no record of any laparoscopic surgery. During the period of follow-up, 844 subjects (93.5%) died, with an overall median survival of 5.2 months (95% CI, 4.9-5.6).

The laparoscopic surgery and no laparoscopic surgery groups were similar with respect to many demographic and clinical characteristics (Table 1). The most striking differences between groups were observed in the proportion of individuals with distant metastases at diagnosis (67.9% of those who had laparoscopic surgery compared with 41.2% of those who did not have laparoscopic surgery; P = .001) and in the proportion who had a surgical biliary or gastric bypass (28.6% of those who had laparoscopic surgery compared with 56.1% of those who did not; P = .001).

The Kaplan-Meier survival curves for persons who had laparoscopic surgery and those who had only open surgery were nearly identical when the entire cohort was evaluated. The median duration of survival in the laparoscopic surgery group was 4.8 months (95% CI, 4.1-6.8) compared with 5.3 months (95% CI, 4.9-5.6) in the group that did not have laparoscopic surgery. The findings were not substantially different when the cohort was stratified into patients without distant metastases at diagnosis and those with distant metastases at diagnosis (Figure 1). Stratification according to the use of surgical bypass also did not demonstrate survival differences.

Unadjusted hazard ratios and 95% CIs for various exposures are presented for the entire cohort as well as with stratification according to the presence of distant metastases at diagnosis and the use of surgical bypass procedures (Table 2). Age, comorbidity, tumor grade, tumor extent, and the use of chemotherapy, radiation therapy, therapeutic endoscopic retrograde cholangiopancreatography, and surgical biliary or gastric bypass were associated with survival. The use of laparoscopic surgery was not adversely associated with survival. The hazard ratio for individuals who had laparoscopic surgery relative to those who did not was 1.01 (95% CI, 0.82-1.24). When analyzed according to subgroups, the point estimate of the hazard ratio for laparoscopic surgery was less than 1.0 (ie, not associated with worse survival), except for the subgroup of individuals who had a biliary or gastric bypass (hazard ratio 1.06; 95% CI, 0.72-1.56).

The consistency of the association between laparoscopic surgery and survival was tested by comparing hazard ratios associated with the use of laparoscopic surgery (compared with not having laparoscopic surgery) within several exposure categories, including age, sex, household income, comorbidity, tumor site, tumor grade, presence of distant metastases at diagnosis, use of radiation therapy, and use of bypass surgery. The rate of death associated with laparoscopic surgery was similar across all of the exposures we examined.

Because imbalances in important determinants of survival may have accounted for some of the variation in the rate of death among subjects, we performed selected multivariate analyses to adjust for various prognostic factors (Table 3). In general, adjustment for the known determinants of survival tended to lower the hazard ratio for laparoscopic surgery (ie, reduced the value toward 0 from 1.0). The CIs around adjusted estimates are wide because data on some covariates were unavailable for many subjects and only individuals who had a complete set of predictors were included in multivariate models. When adjusted for age, sex, tumor characteristics (size, grade, and the presence of lymph node and distant metastases at diagnosis), and the use of other interventions (chemotherapy, radiation therapy, therapeutic endoscopic retrograde cholangiopancreatography, and surgical biliary or gastric bypass), the hazard ratio associated with the use of laparoscopic surgery was 0.93 (95% CI, 0.62-1.40).

Laparoscopic surgery has revolutionized the way surgeons perform many common abdominal procedures. Nevertheless, laparoscopic resection of gastrointestinal neoplasms has not become widespread despite promising reports documenting the feasibility of minimally invasive resection of colorectal^7,19 and gastric²⁰ tumors. Compared with popular procedures, such as laparoscopic cholecystectomy, herniorrhaphy, and fundoplication, the relative ambivalence with which surgeons have embraced laparoscopic cancer surgery is related to many factors. Some of these undoubtedly include the technical complexity of advanced laparoscopic procedures, the lower volume of cases available for achieving competence, and the lack of an unequivocal benefit over conventional surgery such as that associated with laparoscopic cholecystectomy. However, one of the most serious concerns regarding the use of laparoscopic surgery in treating gastrointestinal malignancies has been the fear that exposure to carbon dioxide pneumoperitoneum is harmful to patients with gastrointestinal cancer.³

There is little doubt that some of the disturbing case reports of early port site recurrence and peritoneal dissemination of early-stage cancers were related to poor operative technique and it is now widely recognized that care must be taken when handling and extracting cancer-bearing tissues during laparoscopic surgery.²¹ Still, many investigators have reported that carbon dioxide pneumoperitoneum itself alters the natural history of malignant tumors, although the nature of this relationship varies from study to study.^6,21- 26 The mechanism of any biological effect is poorly understood. Some of the factors implicated in in vitro and in vivo animal studies include aerosolization of tumor cells^23,27 and a direct⁶ effect of carbon dioxide pneumoperitoneum on tumor growth. Other studies have emphasized the importance of the size of the inoculum of tumor cells,²⁸ the temperature of the insufflated gas,²⁹ and traumatization of peritoneal surfaces.³⁰

Our objective in this study was to investigate the effect of laparoscopic surgery on the survival of a homogeneous group of patients with abdominal malignancy. We chose to study patients with pancreatic cancer because these patients frequently have advanced disease at diagnosis and potentially expose a large volume of tumor cells to the environment of a carbon dioxide pneumoperitoneum. Since survival is poor in unresected pancreatic cancer,³¹ many patients would be expected to die during follow up, increasing the probability of detecting a difference in mortality rates in a survival analysis.³²

We did not find that exposure to laparoscopic surgery had an adverse effect on survival in patients with pancreatic cancer who did not have a pancreatic resection. In fact, the point estimates of the relative rate of death in most of our analyses favored the group that had laparoscopic surgery. Our findings do not suggest that the adverse effect of carbon dioxide pneumoperitoneum on cancer growth reported in some laboratory studies is an important phenomenon in humans. The wide CIs around our estimates indicate that this study cannot exclude relatively modest effects (an increase in the relative rate of death of 40% or less). Nevertheless, the upper limits of the 95% CIs do not include values as extreme as would be expected if laparoscopic surgery had a strong effect on cancer behavior. Certainly these data are not consistent with the suggestion of case reports that there is a many-fold increase in the rate of cancer progression associated with laparoscopic surgery.

The findings of our study must be interpreted with caution. We used an observational study design, which is more susceptible to selection bias than randomized studies when comparing interventions. We used data from databases that were not necessarily designed to permit the type of analysis that we have done. The Medicare claims files, for example, use codes for procedures that may be imprecise and may not capture all procedures performed. Specifically, some outpatient interventions, such as the use of chemotherapy and endoscopic retrograde cholangiopancreatography may be underestimated. Important details regarding tumor characteristics in the SEER file were missing for some patients. Because the subjects we analyzed did not undergo a pancreatic resection, there may be differences in the accuracy with which tumor characteristics were determined.

Our use of survival as the main outcome measure may not have detected other important adverse effects of laparoscopic surgery, such as an increase in tumor volume or port site recurrences, which did not directly affect the death rate. Because the survival of this cohort was relatively short, it is possible that an adverse effect of laparoscopic surgery might not have been detected since exposed individuals did not survive long enough to experience the effect of exposure. However, clinical reports³³ and laboratory studies of adverse events associated with laparoscopic cancer surgery have documented that they occur in the early postoperative period.

Many sources of measurement error in our analysis, such as imprecision in estimating the day of diagnosis, would likely be nondifferential with respect to whether individuals had laparoscopic surgery. This is of concern since this type of error would be expected to bias estimates of risk toward the null hypothesis, and here we report the lack of an association. Exposure misclassification is possible if patients who undergo a laparoscopic procedure and are "converted" to an open procedure are coded only as having undergone an open procedure. We do not think that this was a major source of error in our analysis because many patients had records for both a laparoscopic procedure and a laparotomy, even if the 2 occurred on the same day. Finally, it is of concern that many patients who had laparoscopic surgery also had open surgery, either at the same time or at a different time. However, our main goal in this study was to evaluate the effect of exposure to laparoscopic surgery on cancer survival, irrespective of other interventions. Our findings should not be interpreted as a direct comparison of laparoscopic surgery with open surgery.

We believe that our findings are valid for several reasons. Because all of the subjects in the cohort had 1 of 2 procedures that are equivalent in their indications, timing, and effectiveness,³⁴ the effect of selection bias would be diminished when compared with observational studies that compared more heterogeneous interventions. In fact, it seemed that patients who were exposed to laparoscopic surgery had a poorer baseline prognosis than those who were not, as evidenced by the presence of distant metastases at diagnosis (one of the most important prognostic factors in pancreatic cancer).^31,35- 36 This suggests that any selection bias was toward offering laparoscopic surgery to patients who were at higher risk, adding further support to our finding that survival in the laparoscopic surgery group was at least as good as in the alternative group.

We adjusted survival estimates to account for confounding by other determinants of prognosis. We also excluded patients who had any type of pancreatic resection (or ablative procedure) or cholecystectomy. Pancreatic resection was used as an exclusion criterion to create a more homogeneous cohort. Patients who had a cholecystectomy were excluded because there likely are systematic differences in the timing of the diagnosis of cancer depending on whether an open or laparoscopic operation was done in a patient who happened to have a pancreatic neoplasm. Since it may be more likely for pancreatic cancer to be discovered intraoperatively during an open cholecystectomy compared with a laparoscopic cholecystectomy, the survival rate from diagnosis may seem spuriously longer in patients who had open cholecystectomy ("lead time bias").

Our findings require corroboration. Long-term results of randomized trials comparing laparoscopic with open resection of colorectal cancer^7- 9 will provide further insight into the oncologic efficacy of laparoscopic cancer surgery. Although randomized trials will provide a higher level of evidence than the observational methods that we have used, it may take several years for adequate follow-up data to be available.

In conclusion, exposure to laparoscopic surgery did not have an important adverse effect on survival in a cohort of elderly patients with pancreatic cancer who had laparoscopic surgery or a laparotomy but no pancreatic resection. We did not find evidence that the effect of carbon dioxide pneumoperitoneum on promoting neoplastic growth reported in animal studies is an important mechanism in human cancer.

Support for this research was provided by the Legacy Good Samaritan Hospital Foundation, Portland, Ore.

We are indebted to Louis Homer, MD, PhD, for his help in developing the study protocol; to Olukemi Quinn and Cindy Muller for their expert assistance with literature searches and citation retrieval; and to John Birkmeyer, MD, Joan Warren, PhD, Robert Glasgow, MD, and Stacey Urbach, MD, MPH, for thoughtful comments on earlier versions of the manuscript. This study used the linked SEER-Medicare database. The interpretation and reporting of these data are the sole responsibility of the authors. The authors acknowledge the efforts of the Applied Research Branch, Division of Cancer Prevention and Population Science, National Cancer Institute, Bethesda, Md; the Office of Information Services and the Office of Strategic Planning, Health Care Financing Administration, Washington, DC; Information Management Systems Inc, Silver Spring, Md; and the SEER Program tumor registries in the creation of the SEER-Medicare database.

Corresponding author and reprints: David R. Urbach, MD, Toronto Western Hospital, 399 Bathurst St, Room MP8-332, Toronto, Ontario M5T 2S8, Canada (e-mail: David.Urbach@uhn.on.ca).