Predicting Readmission and Leak Rates Following Colectomy with Red Cell Distribution Width: A Simple but Effective Test

Results: A total of 118 patients underwent colectomy during this period. Readmission and/or anastomotic leak occurred in 49 patients. The sensitivity of elevated RDW levels (greater than or equal to 14.0) at detecting future readmission and/or leak was 89.8%. The negative predictive value for a normalized RDW below 14.0 at predicting the non-occurrence of leak or readmission was 87.7%. The specifi city of an elevated RDW was 72.4% and the positive predictive value was 76.5%.


Introduction
Quality metrics are increasingly being scrutinized as part of today's health care environment. Two important quality metrics include length of stay and readmission rates. Colorectal surgery is known to have one of the highest morbidity rates, accounting for a disproportionate amount of complications compared with other procedures [1]. Readmission rates following colorectal surgery can be 20% or more [2][3][4][5][6]. The ability to predict some of these complications has been an elusive goal. Recent studies show that only a small proportion of readmissions could be predicted [7]. More importantly, pay for performance measures including readmission rates are currently used to drive reimbursement [8][9]. Therefore the ability to predict adverse events would be very helpful in assessing risk, as this can be used to adjust expected versus actual morbidity following surgery. Many studies have assessed infl ammatory measures as potential predictors of anastomotic leak and readmission. In particular, there is some evidence that C-reactive protein (CRP), an acute phase reactant that is released in response to ongoing infl ammation, can be predictive of leak or readmission following colorectal surgery [10][11]. None, however, have been suffi ciently sensitive to withhold discharge compared to standard clinical assessment.
We hypothesized that red cell distribution width (RDW) can be used as an alternative measure of ongoing nonspecifi c infl ammation, and this may predict future morbidity.
RDW is a measure in size variation of circulating red blood cells. Increased size (anisocytosis) can represent a variety of conditions, including nutritional defi ciency, various types of anemia, hemoglobinopathies and it is a non-specifi c marker of infl ammation. It has previously been used to predict mortality, though its ability to predict other adverse events remains uncertain.

Methods
A retrospective review of patients admitted to a large, urban safety-net hospital over a two year period from 2010-2012. An administrative database kept within the division of all procedures performed was reviewed after Institutional Review Board approval was obtained. Variables obtained from a retrospective chart review included age, gender, surgical procedure, length of hospital stay (LOS), values of RDW at all time periods, the incidence of anastomotic leak, and readmission rates within 30 days following hospital discharge. While labs are typically obtained routinely on day 1 and 2 postoperatively (and include RDW), any laboratory values obtained later than this were deemed clinically necessary and not done as part of a standardized protocol.
A cutoff of 14.0 was used as the upper limit of normal for RDW. The reference range for our laboratory was 9-14%.
Positive and negative predictive values as well as sensitivity and specifi city were calculated (as defi ned in Table 2). Inclusion criteria included any partial colectomy with an anastomosis > 10 cm from the anal verge. Anastomotic leak was defi ned as contrast extravasation or peri-anastomotic fl uid or gas on radiographic imaging or breakdown of the anastomosis seen at operative re-exploration. All surgeries were performed by a colorectal surgery resident or general surgery chief resident with the help of one of 5 board certifi ed colorectal surgeons.
All patients were managed postoperatively with an enhanced recovery protocol that has previously been published [12].
Discharge criteria included tolerance of a diet without need for supplemental intravenous fl uids or intravenous narcotics, evidence of bowel function (fl atus, bowel movement, or similar ostomy output), and no clinical concern for an infectious process such as anastomotic leak. Emergent cases were excluded.

Results
A total of 118 patients underwent colectomy during this period. Clinical characteristics can be found in Table   1. Fifty eight percent were male with an average age of 58.4±14.6. Average ASA was 2.9±0.6. Average LOS was 7.2±  Table   2). The specifi city of an elevated RDW was 72.4% and the positive predictive value was 76.5%. Reasons for readmission are characterized in Table 3. There was no difference in readmission rates among various groups at high risk for alterations in RDW (Table 4). Positive predictive value for each individual event was lower: 42% for readmission, 48% for leak.

Discussion
The ability to predict (and therefore possibly prevent) complications such as readmission and anastomotic leak following surgery has been an elusive goal. Recent studies show that only a small proportion (21%) of surgical readmissions could be predicted [7].. Dehydration and infection were some of the more commonly cited areas of preventable readmissionsareas that we also saw as common reasons for admission. Thus, the ability to predict and therefore more closely scrutinize these patients may be a useful tool, particularly when pay for performance measures will start to use readmission rates as benchmarks. This is particularly true in tertiary care or safety net hospitals such as our own, where our previously reported results on complex patients suggest this is a diffi cult problem [9,13,14]. Therefore the ability to accurately adjust expected versus actual morbidity following surgery will become increasingly important, and there are limited tools for this type of evaluation. Our data suggests that RDW may be one tool to effectively do that.
There have been other studies, specifi cally regarding C-reactive protein (CRP) to try and achieve similar predictive models after colectomy. Krapta et al [15]. Demonstrated that non-elevation of CRP was able to accurately predict shorter length of hospital stay (p<0.01) with average CRP values of 6.3 in those with a LOS < 3 days, vs 11.7 in those with LOS > 4 days. Although the CRP values were different in readmitted vs. non-readmitted patients (11.8 vs. 9.9), the difference did not reach statistical signifi cance. Despite the small sample size, their early data have given promise to the ability to predict adverse outcomes, and thus risk ASA: American Society of Anesthesiologist's Score *Of the 40 patients with cancer as indication for surgery **Defi ned as resection to level below peritoneal refl ection but above 10 cm adjust predictive models in an era of pay for performance.
Others have also investigated CRP as a predictor of safe, early discharge without subsequent complication. Giaccaglia et al [16]. Demonstrated a >96% negative predictive value for leak when the CRP was <16.9mg/mL on postoperative day 3 and >98% when CRP was <12.5mg/mL on day 5. While they suggested that procalcitonin may enhance this negative predictive effect, it is an expensive test [17], and not routinely ordered. IL-6 is another non-specifi c infl ammatory marker that also has similar (albeit lower) predictive value for subsequent complications [18]. Ortega-Deballon et al [19] also showed that CRP (cutoff of 125 mg/L on postoperative day 4) had a negative predictive value of 95.8%. There is also a meta-analysis of 7 studies that demonstrated CRP was a useful negative predictive test for leak [20].
With these types of serum biomarkers for adverse events, authors have shown that a positive predictor of leak is a much more diffi cult entity. Our results suggest similar fi ndings, in that positive predictive values overall were 76.5%, but were much lower for an individual event (42% for readmission, 48% for leak). No single biomarker has been shown to be effective to accurately predict a positive occurrence of a complication.
However, negative prediction seems to be more feasible [21].
These serum biomarkers are not part of a routine order set, so we sought to characterize something that is readily available on existing tests. The role of RDW as related predictor has not been well characterized in colorectal surgery. However, there is some precedent that RDW values are a non-specifi c marker for ongoing infl ammation and thus can potentially be used as a predictor of outcomes. It has been used to predict mortality in many medical patients [22][23][24][25], with hazard ratios for increased risk of death ranging from 1.09-1.31 for each 1% increase in RDW values. These differences in mortality persisted even after adjusting for severity using APACHE scores and including comorbid conditions such as renal failure, respiratory failure, and other potential confounding factors. It has been shown to potentially be superior to C-reactive protein (CRP) as a predictor of adverse outcome [26]. The true value of RDW in surgical patients has been incompletely studied. Our study suggests, as seen in the medical studies, that it may be helpful in predicting adverse outcomes.
In acute pancreatitis, Senol et al. [27] reported that an elevated RDW above 14.8 as part of a predictive model correctly predicted adverse outcome in 77% of cases. Elevated RDW has also been used to predict ongoing disease activity in ulcerative colitis and Crohn's disease patients [28][29]. These studies suggest that RDW can be used as an active predictor of ongoing infl ammation, which could signify an impending problem such as anastomotic leak, abscess, or other problem that would require readmission. A study in hernia patients further substantiated this claim [30]. The ability to predict ongoing infl ammation may not be as accurate in obese patients [31].
It has been established that an elevated RDW is a potential indication for colonoscopy. In one study, RDW was 84% sensitive and 88% specifi c for right sided colon cancer [32].
in appropriately selected patients. These numbers parallel our fi ndings and further suggest the potential role for RDW of an ongoing infl ammatory process.
There is much more limited data on surgical patients.
Warwick et al. [33]. showed similar results to our study in lung surgical patients. RDW was a signifi cant predictor in need for mechanical ventilation, LOS, in-hospital and long term mortality. Similar results were observed by Polat et al [34], who found that elevated RDW values were predictive of ICU and overall LOS as well as mortality in pediatric heart surgical patients. To our knowledge, this is the fi rst application to colectomy patients, and the results suggest this could become a useful tool for predicting adverse outcomes. Our data are surprising in that such a simple test can be used to accurately predict adverse outcome, with sensitivity of 89.8%.
Furthermore, the lack of elevation (negative predictive value for a normalized RDW) is also comforting in that it too predicts lack of adverse outcome (negative predictive value in our data = 87.7%). The test is not as specifi c, in that a number of other things may cause this elevation.
Several aspects of our study deserve some additional scrutiny. We describe a relatively high leak rate (23.7%) compared to reported data. This is likely a result of multiple factors, including the fact that our patient population is at high risk based upon advanced stage of disease, urgency of procedure, and nutritional deprivation. These are a byproduct of working at a large urban safety-net hospital.
Of all surgeries, 46 were performed for cancer. Early T-stage Negative predictive value if RDW <14** 87.7% *As defi ned as those with RDW>14 with leak or readmission (true positive) / (true positives +those with leak or readmission and RDW < 14). **Defi ned as those without leak or readmission and RDW < 14 (true negatives) / (true negatives + readmission or leak with RDW < 14).  describing how infectious and other complications can be increased in our population [9,14]. This also likely contributed to increased length of stay ( Complications and anastomotic leak rates were quite high, likely as a result of our complex patient population with advanced disease and multiple risk factors for complications [9,14]. This could bias the results and make them less applicable in other scenarios, as our patients may have had a higher pretest probability for detecting problems. For this reason, we also include the negative predictive value, which is also insightful and as pointed out earlier, probably more feasable than positive predictive values (which have proven remarkably elusive).
Finally, RDW is a non-specifi c laboratory value and the clinical implications of an elevated value are uncertain (as it could represent a wide range of etiologies). This problem has also been seen with other non-specifi c infl ammatory markers (e.g. CRP). Similarly, the accuracy of either occurrence alone (e.g. leak alone or readmission alone), is not as good as a composite using both endpoints. This is because there is a process going on that may give a false representation of what is going on if only one is investigated. Despite this, the use of RDW is a readily attainable factor that can be useful in predicting readmission and leak rates following major colorectal surgery.
Our results show some initial promise regarding this approach and further study is needed. It remains unclear whether the threshold for discharge should be higher in patients who do not have RDW<14, but our study suggests they should perhaps have closer follow-up. Larger, more prospective and randomized studies will be required to confi rm these fi ndings.