Introduction
Over the last three decades, the operative mortality and lengths of stay have decreased following pancreatic resection,
1‐
6 which can be attributed to increasing regionalization of care,
7‐
10 improved perioperative and critical care,
11‐
13 improved prevention and management of complications, implementation of critical pathways,
14,
15 and improved post-hospital inpatient and outpatient care. Despite the improvements in mortality and lengths of stay, the morbidity rates, usually defined as the occurrence of any complication in the postoperative period, remain high with reported rates in excess of 30% even at major centers.
3,
4,
6,
16‐
20 Readmission, a good measure of morbidity, is rarely reported. In addition, when reported, the focus is on readmissions within the first year.
There are three previous studies evaluating readmission following pancreatic surgery.
16,
21,
22 Two studies are single-institution studies, both of which included pancreaticoduodenectomy for benign and malignant disease.
16,
21 Neither study reported readmissions within 30 days of discharge. Emick and colleagues
16 reported a 19% readmission rate in the year after surgery in 1,643 patients undergoing pancreaticoduodenectomy. van Geenen and colleagues
21 reported an overall 1 year readmission rate of 38% in 283 patients undergoing pancreaticoduodenectomy. Given the single-institution nature of these studies, readmissions to other facilities may not be identified, so the reported rates may not reflect national readmission rates.
A population-based study using the California tumor registry and hospital discharge data reports a 59% readmission rate in the year after pancreaticoduodenectomy in patients with pancreatic cancer.
22 They also report decreased long-term survival in the group requiring readmission. The majority of readmissions were related to disease progression. As such, they are a marker of early mortality and not the cause. None of the above studies evaluated readmissions using a time-to-event analysis and therefore potentially underestimated readmission rates.
The goals of our study were to use a population-based data set [Surveillance, Epidemiology, and End Results (SEER)–Medicare-linked data] to evaluate the readmission rates using time-to-event methods. We also evaluate the reasons for readmission within 30 days of pancreatic resection (early readmission) and between 30 days and 1 year (late readmissions). We hypothesize that early readmissions are related to operative complications, contribute to early mortality, and are potentially preventable. Conversely, late readmissions are associated with disease progression and are a marker, rather than a cause, of early mortality. Therefore, it is critical to analyze 30-day readmissions separately. We also determine the patient and tumor factors associated with early readmission and perform a survival analysis to determine the effect of early and late readmission on survival.
Discussion
Our study demonstrates an overall readmission rate of 53% and an early (within 30 days) readmission rate of 16% after pancreatectomy for pancreatic cancer in Medicare patients. The overall readmission rate, calculated using at time-to-event analysis, is similar to the 59% readmission rate in a previous population-based study
22 and higher than the reported rates in previous single-institution studies.
16,
21 The higher rates in population-based studies more likely represent true readmission rates in the general population. Moreover, Yermilov and colleagues
22 found that 47% of readmissions were not to the hospital performing the primary surgery. It is likely that the single-institution studies did not capture readmissions to outside hospitals and may grossly underestimate readmission rates even in their own patients. In addition, the non-time-dependent methods used in previous studies will inflate the denominator or number at risk in a given period, decreasing the observed readmission rates.
As hypothesized, the length of the time period elapsed since undergoing pancreatectomy determined the reason for readmission. Early readmissions were more commonly associated with postoperative complications, while late readmission after 30 days was more likely to be due to disease progression (metastases or recurrence).
Dehydration occurred in approximately one quarter of patients in both the early and late groups. When evaluating diagnosis codes concurrent with dehydration, dehydration was more commonly related to surgical complications in the early readmission group while dehydration in the late group was more commonly related to chemotherapy or recurrence of pancreatic cancer and general failure to thrive. Likewise, the nature of postoperative complications differed between the early and late groups. Postoperative complications requiring early readmission most commonly included sepsis, abscess, anastomotic leak, and acute hemorrhage, whereas late complications included small bowel obstruction, incisional hernias, biliary strictures, and cholangitis. A diagnosis of delayed gastric emptying or gastric outlet obstruction was seen in 8% of early readmissions and 4% of late readmissions. Similar to dehydration, the reasons for the delayed gastric emptying or gastric outlet obstruction differed between the early and late groups. Early delayed gastric emptying following pancreaticoduodenectomy has been reported in 10–20% of patients immediately following pancreatic resection
3,
25 and accounts for the majority of delayed gastric emptying or gastric outlet obstruction in the early group. In the late group, however, this diagnosis was associated with gastric outlet obstruction secondary to tumor recurrence.
Only the initial length of stay and the type of resection predicted early readmission. Those who had an initial length of stay of ten or more days were more likely to require early readmission. A diagnosis code for operative complications did not predict readmission. This suggests two things: first, not all operative complications are noted during initial admission and second, not all complications lead to readmission. A prolonged initial length of stay does not cause readmission; rather, it is likely a marker of serious postoperative complications, the most common diagnosis during early readmission. In addition, longer lengths of stay predispose patients to developing additional iatrogenic infections, as well as VTE/PE and atelectasis associated with prolonged immobility, which typically occur in a hospital setting. There was no association of age or patient comorbidities with early readmission.
This is the first study to demonstrate that patients undergoing distal pancreatectomy have an increased risk of readmission. This is unexpected as pancreaticoduodenectomy is a more complex procedure and thought to be fraught with more complications. However, pancreatic fistula rates have been reported to be higher following distal pancreatectomy than pancreatic head resection.
26‐
28 This fact, coupled with the fact that distal pancreatectomy is less likely to be performed at high-volume centers by experienced surgeons,
8 likely contribute to this finding.
In the first year after initial discharge, deaths due to pancreatic cancer became a significant competing event. Considering deaths as censored creates informative censoring, since the same factors that influence cancer deaths, likely influence late readmissions. As a result, the multivariate model evaluating the factors associated with late readmission, which treated deaths as censored, does not demonstrate the same predictive factors as the model that treats death as a competing event. By treating deaths a censored, patients with advanced tumor stage (distant disease and positive nodes) are removed from the at risk cohort. However, these factors are related to recurrence, the most common reason for readmission, and would likely have led to readmission in the absence of death.
While the median survival was lower in patients requiring early readmission compared to those who did not, the long-term survival was identical at 18%, suggesting that operative complications increase early deaths. However, survivors of these complications can expect similar survival to their counterparts who had an uncomplicated postoperative course. Late readmission is more commonly due to recurrence and is a marker of early mortality. As expected, it was associated with significantly worse median and long-term survival as shown previously.
22
This study has several limitations, mostly related to the use of administrative data. The reported reasons for readmission in Table
3 were based on identification of specific ICD-9 diagnosis codes both in the primary discharge diagnosis and additional diagnoses provided for the same discharge. We also individually reviewed each readmission record and looked at the diagnosis and procedure codes and gave each readmission a primary reason for the admission. The results were similar using the two methods, in that early readmissions were related to surgical complications and late readmissions were related to recurrence; however, these were subject to the reviewers’ interpretation.
It is often difficult to identify specific complications commonly reported after pancreatic surgery using administrative data, including pancreatic fistula and bile leak. For example, there are codes for postoperative complications and anastomotic leak, but they are not specific. In addition, the administrative data is used for billing purposes, so diagnosis codes mandating reimbursement may be more likely to be coded. While we were able to look at nodal status, data were not available on margin status to evaluate its effect on early and late readmission.
In summary, this study demonstrates the rates and the most frequent causes of early and late readmissions and identifies predictors of hospitalization during these time periods after initial discharge following pancreatectomy for pancreatic cancer. These findings reinforce the finding that readmission rates in the general population following pancreatectomy occur in over 50% of patients and are underreported in single-institution studies. Additionally, this study delineates the factors contributing to early and late readmissions. It demonstrates that early readmission related to complications shortens median but does not affect long-term survival if the patient survives the operative complication. Late readmissions are a marker of early mortality. Death due to cancer is a competing event with late readmission. As such, the factors influencing late readmission are similar to those that predict early mortality. The 15% of readmissions related to operative complications are, therefore, potentially preventable. The reasons for early readmissions need to be studied further to identify individual factors and operative techniques that decrease these preventable readmissions.
Deepthi Martha Reddy, Presenter (University of Texas, Galveston, TX medical student)
Discussant
Dr. Sharon Weber (Madison): First, I have to congratulate you as a medical student in presenting this work. This is a very timely paper and I am really happy to see it presented here at the SSAT. As many of you know, CMS plans to use readmission as a quality of care indicator in the future because the estimated cost of readmissions has been estimated at about $17 billion. We know we have underestimated the rate of readmission when utilizing single-institution studies because of readmission at other hospitals. Using the SEER-Medicare database is a great way to obtain the actual rate of readmission, so I congratulate you on this work.
I have questions surrounding two main points. First, “how can we impact this?” and the second question surrounding this issue—“Are these findings real?”
To address the first question, clearly, the mortality of almost 8% in-house and 23% at 30 days is not acceptable. In addition, the readmission rate of 16% at 30 days is also very high, considering that the median length of stay was 14 days.
Your group has presented some of the seminal work looking at hospital volume, and I am wondering if you did not look at that here. Is hospital volume one area where we may be able to impact the rate of readmission and mortality? Were high-volume hospitals less likely to have higher numbers of readmissions? In addition, was there any difference in geographic patterns for readmissions?
Secondly, a recent publication by Coleman, in the New England Journal in April 2009, examined readmissions for Medicare patients using claims data. Of those 800,000 patients who underwent both small and larger surgical procedures, the readmission rates at 30 days and 1 year were almost identical to yours. Thus, this leads to the question, “do your findings represent a real phenomenon—that the readmission is higher after pancreatectomy, which is clearly a more complex operation than the average surgical procedure?” Or do these findings just imply that the Medicare population has a higher rate of readmission overall, perhaps because of increased age?
Closing discussant
Deepthi Martha Reddy: Thank you, Dr. Weber. We did not include data on hospital volume. We did not do so because some of the hospitals at which patients might go to undergo pancreatic resection may not be included in the SEER regions. As a result, they may falsely appear as low-volume hospitals.
When we evaluated hospital volume excluding hospitals not in SEER regions, hospital volume predicted mortality, but not readmission.
For the second question, I would like to refer to my mentor Dr. Riall.
Closing discussant
Dr. Taylor S. Riall (Galveston, TX): We did not include hospital volume because some of the hospitals at which patients might undergo pancreatic resection may not be included in the SEER regions. As a result, they may falsely appear as low-volume hospitals. For instance, Johns Hopkins is a high volume hospital, but it is not in a SEER region.
If you take patients who live in New Jersey, which is a SEER region, they may travel to Baltimore to have their surgery done at Johns Hopkins. In the database, we would be able to identify Johns Hopkins as an individual hospital, but it would not appear as a high volume hospital, since we would be calculating volume based only on the number of patients living in SEER regions who had their surgery done there.
In addition, you are looking at Medicare volume and not total pancreatic resection volume. Therefore, there are inherent problems with looking at hospital volume. When we evaluated hospital volume including only hospitals in the SEER regions, hospital volume was a predictor of mortality and increased length of stay, but not readmission.
With regard to your question regarding Medicare readmission rates compared to readmission rates for pancreatectomy specifically, I think you make a good observation. Even in the single institution studies, the readmission rates are high. Therefore, I think this is actually real and not simply the readmission rates for the Medicare population. Readmission is common following pancreatectomy, and we need to evaluate the reasons for readmission and areas for improvement in a multi-institutional setting. This can be increasingly important in this pay-for-performance era.
Discussant
Dr. Keith D. Lillemoe (Indianapolis): Again, I would just echo that the medical students here put us all to shame. Great presentation.
There is a bit of a disconnection. The 8.3%, 30-day mortality is high. Obviously it is not a high-volume center, tertiary center, or teaching hospital. It is a national database. Regardless, it is still too high.
However, the 15% readmission rate is very acceptable. Why is there such a disconnection? Is it the fact that these people are dying before they get readmitted? I do not quite understand your data because 15% readmission is about as good as you are going to see from any of the best of institutions, whereas an 8.3% mortality is unacceptably high.
Could you explain that disconnection to me? Is it something related to the data analysis or the database that you are using?
Closing discussant
Deepthi Martha Reddy: The early the 15% readmission rate was related to postoperative complications. This rate is likely lower than reported readmission rates since most studies look at 1 year readmission rates and not 30-day readmission rates.
Discussant
Dr. Keith Lillemoe (Indianapolis): However, do you not anticipate that the 8.3% operative mortalities are dying of postoperative complications? Those are not tumor progression for Whipples or pancreatic resections in 30 days.
Discussant
Dr. Charles Vollmer (Boston): I would like to shift gears and take it from the administrative level back down to the practice level. And the one thing that really struck me was the fact that if you are in the hospital for greater than 10 days, you have a very high chance of being readmitted soon thereafter.
These are cases where there is a deviation in the standard progression of the postoperative recovery period. And I wonder if we as surgeons can find a way to impact on that readmission rate by figuring out what we are doing wrong, or what is going on with the patient, in that first 10-day period or first stay.
Therefore, in other words, what could be predictive factors from the in-house recovery period that would say this person should not be sent home at this point? Maybe we are doing a disservice in trying to cut the length of stay days down, on some of these patients when we could tidy them up and solve the problems by keeping them in the hospital longer. Any thoughts?
Closing discussant
Dr. Taylor S. Riall: I personally think we are seeing these readmissions when we do not recognize postoperative complications. When you look at the readmission rates before and after initiation of critical pathways, you see decreasing length of stay and the readmission rates actually go down.
Therefore, I do not think the answer is to keep those people there longer to prevent the complications but, as you suggest, to identify the ones who have occult problems and need to stay. I think the patients that get readmitted are the occult complications that we do not recognize. For example, we might miss a pancreatic fistula that did not show up in the drains, so the patient appears to be “on the pathway.” Then, we send them home, and they develop an abscess. I am not sure we are going to be able to reduce our readmission rates to zero, but I think it would be beneficial for high-volume centers to pool our data and identify factors predictive of readmission. This could potentially cut down readmissions and cost significantly.
I think one way to do it is to continue to centralize pancreatic resection at high volume centers. We could incorporate these predictive factors into our pathways.
Discussant
Dr. Henry Pitt (Indianapolis, IN): We have the NSQIP data from 2005–2007 on 2,000 pancreatectomies, and the mortality is less than 3% in that data base.
Closing discussant
Dr. Taylor S. Riall: This is Medicare data, and I suspect the higher observed operative mortality is expected. Increased mortality following pancreatic surgery in elderly patients has been well documented, so I would expect a higher rate in this data set than NSQIP, which includes patients of all ages and resections done for benign disease.