Centering health care delivery on providing and measuring value has gained increased prominence in recent years. Although defining value is much more complex than simply reducing costs,1 unplanned hospital readmissions adversely affect the value of the delivered care. Interest in hospital readmissions has grown, as it was reported that unplanned readmissions cost Medicare $17.4 billion in 2004 and $26 billion in 2013.2, 3 The 30-day hospital readmission rate is a measure that is now being used to compare hospitals in a publicly available manner,4 and hospitals with excessive readmissions have been punished with reimbursement denials and reductions for Medicare patients through the Medicare Hospital Readmission Reduction Program.5, 6
Over the past 4 decades, the annual number of spinal surgeries has increased, due to an expanding pool of elderly patients with spinal pathology, improved understanding of various disease processes, introduction of less-invasive techniques, and the utilization of more advanced technologies. Along with this increase in utilization has been a significant increase in cost associated with spinal surgery.7 Due to the high cost of surgery and the morbidity associated with complications, there is a significant interest in minimizing the incidence of readmissions.
Recent work has examined predictors of readmission after spinal surgery across a wide variety of patients at a tertiary referral center8 as well as predictors of readmissions for specific spinal procedures.4, 8, 9 However, more hospital-wide data are needed to establish clinically relevant readmission rates to serve as a representative benchmark for future work and comparisons. Our study evaluates the 30-day readmission rate at a single institution, determines how many of these were procedure-related readmissions, and identifies specific risk factors that independently predicted a higher likelihood of readmission within 30 days.
MATERIALS AND METHODS
After obtaining IRB approval, data from 1187 consecutive spinal surgery admissions on 1137 unique patients between January 1, 2010, and December 31, 2010, were obtained using the Northwestern University Electronic Data Warehouse (EDW), an initiative funded jointly by Northwestern University Feinberg School of Medicine, Northwestern Medical Faculty Foundation (NMFF), and Northwestern Memorial Hospital (NMH) (NUCATS grant UL1RR025741). Current procedural technology (CPT) codes were used to identify spinal procedures. Of spine surgeries identified by CPT codes, all were included for analysis, except for spinal cord stimulator trials and fiducial placements. As some patients underwent multiple operations, each surgical encounter was analyzed, rather than each patient. In accordance with current Center for Medicare & Medicaid Services (CMS) guidelines, a readmission was defined to be an inpatient stay at Northwestern Memorial Hospital for any reason within 30 days of discharge from the initial hospital stay.10 Thus, emergency department visits, observation visits, and outpatient surgeries were not considered to be readmissions.
For each patient admission, characteristics of both the patient and procedure were acquired. Patient information (Table 1) included age at surgery, BMI, gender, smoking status (never, >12 months ago, <12 months ago), insurance type (workers compensation, public, or private), and whether a comorbid condition was present (cancer, cardiac, endocrine, respiratory, or renal), as identified by the International Classification of Disease, ninth edition (ICD-9). Spine-related diagnoses were identified from individual patient records. Indications for spinal surgery were classified into degenerative disc disease, spinal stenosis, intradural lesion, metastasis, trauma, scoliosis or other spinal deformities, and prior hardware failure. Operative characteristics included whether a fusion was part of the procedure, the length of surgery (minutes), number of surgical levels, and whether the procedure was deemed “Minimally-Invasive” (MIS) or “Open” by attending surgeons.11 After the procedure, total time (hours) spent in the intensive care unit (ICU) and length of stay (days) for the entire hospitalization were recorded.
To determine the clinically relevant readmission rate, readmissions were classified as procedure related or procedure unrelated in accordance with accepted definitions.12 Examples of procedure-related readmissions included nosocomial infections, falls, SSIs, VTEs such as deep venous thromboses and pulmonary emboli, postprocedure complications, and CSF leaks. Examples of procedure-unrelated readmissions included unrelated exacerbations of preexisting chronic conditions, new conditions resulting from trauma, or unrelated community infections. Two attending neurosurgeons who were not on staff at our institution during the time period of the study (Z.A.S., N.S.D.) performed chart reviews and applied the criteria. Fleiss’ Kappa was used to assess the degree of interobserver agreement between these 2 reviewers, with P value less than 0.05 used to denote significance. In order to classify each case as procedure related or procedure unrelated, in cases wherein there was disagreement, ties were broken by a third attending neurosurgeon (R.G.F.). This neurosurgeon was on staff during the year of operations studied, but did not operate on any patients upon which there was disagreement between the other 2 reviewers.
A series of generalized linear models were constructed with logit link for readmission (yes/no) within 30 days using each of the following variables as predictors: age, gender, smoking status (yes/no within the last 12 months), comorbidities (including cancer, cardiac condition, endocrine condition, respiratory condition, and renal condition), fusion, length of surgery, number of surgical levels, invasiveness (open/MIS), time in ICU (hours), and LOS (length of stay, days).
We used generalized estimating equations (GEEs) to estimate parameters and model-based (naive) standard errors for inferences in order to account for the cases of multiple observations per patient (there were a total of 45 patients with multiple surgeries in the dataset, and 5 of these subjects had 3 surgeries).
Following these bivariate analyses, we determined a list of candidate predictors of 30-day readmission and developed an overall predictive model for the outcome of readmission within 30 days using a manual backward stepwise selection process. Any variables significantly related to the outcome at the 10% level in the bivariate analyses were considered as potential predictors in the overall maximal model. Different types of comorbidities were condensed into a single indicator variable for presence of greater than or equal to 2 comorbidities. Due to the association between LOS, surgery time, number of surgical levels, we chose the number of surgical levels to represent this group of variables in the initial maximal model. Number of surgical levels was deemed the most relevant because it is determined preoperatively, whereas LOS, surgery time, and time in the ICU were more dependent on intraoperative and postoperative factors. Therefore, the initial predictive model for 30-day readmission included age, BMI, invasiveness (MIS or open), the indicator variable for at least 2 comorbidities, and number of surgical levels. The model-building procedure involved sequential removal of variables from the maximal model one-at-time according to the following criteria: a model type III Wald test significance at the 10% level and the smallest quasi information criterion (QIC, a relative measure of model fit). All analyses were conducted in SAS, version 9.4 (Cary, NC; 2012).
A total of 1187 encounters met the study inclusion criteria. Table 1 summarizes the patient encounter characteristics. Of these encounters, 72 (6.1%) were readmitted within 30 days. Procedures involving the cervical spine (286 procedures) had a readmission rate of 4.2% (12 readmissions), while procedures involving the lumbar spine (796 procedures) had a readmission rate of 6.3% (50 readmissions). Procedures using a minimally invasive technique (224 procedures) had a readmission rate of 3.1% (7 readmissions), and those using an open technique (963 procedures) had a readmission rate of 6.7% (65 readmissions). Surgeons had 15.6 ± 9.5 years (mean ± standard deviation) of operating experience after the completion of residency or fellowship. The most common indications for surgery were degenerative disc disease (82.6%), spinal stenosis (25.2%), and prior hardware failure (12.4%).
On the basis of bivariate analysis, patients who were readmitted were more likely to be older (mean 57.7 vs. 52.7), to have a higher average body mass index (mean 30.3 vs. 28.4), and to have more cardiac, renal, respiratory, endocrine, or cancer morbidities. In addition, in our sample, patients who were readmitted had procedures that involved more spinal levels (median 3.9 vs. 2.0) and, consequently, longer surgical times (median 219.5 vs. 128.5 minutes). Finally, readmitted patients were more likely to be admitted to the ICU during the initial hospital stay (55.6% vs. 18.4%) and to have a longer total postoperative hospital stay (mean 9.9 vs. 3.5 days). Importantly, smoking within the past 12 months, surgery involving a fusion, and minimally invasive surgery were not shown to be associated with an increased rate of readmission (Table 2).
Multivariable analysis revealed only 3 significant predictors of readmission when adjusting for all other potential risk factors: a patient with 2 or more comorbidities [odds ratio (OR) 3.72, 95% confidence interval (95% CI) 1.62–8.56], admission to the ICU (OR 2.68, 95% CI 1.45–4.95), and each additional spinal level involved (OR 1.11, 95% CI 1.02–1.21).
The cause of each readmission is summarized in Table 3. Using the criteria proposed by Shah,12 27 of the 76 (37.5%) of readmissions were considered procedure related. Inter-observer agreement between the 2 attending neurosurgeons who reviewed all cases yielded significant agreement, with a Fleiss’ Kappa score of 0.85 (P < 0.001).
In our study, we determined that the 30-day all-cause readmission rate for spinal surgeries at Northwestern Memorial Hospital was 6.1%. Of these readmissions, 37.5% were deemed procedure related upon attending review, leading to a procedure-related readmission rate of 2.3%. After multivariate analysis, the factors that increased patients’ risk of 30-day readmission were the presence of 2 or more comorbidities, an admission to the ICU, and each additional spinal level involved in surgery.
Our overall readmission rate is consistent with existing literature. Amin et al,8 who also studied all surgeries at a tertiary institution, found a 30-day readmission rate of 5.7%, Wang et al13 examined Medicare claims data and found a nationwide readmission rate of 7.9% following cervical surgery and 7.3% following lumbar surgery (Table 4). Lovecchio et al,9 Pugely et al,14 and Kim et al15 all found a slightly lower rate of readmission, but these authors studied only individual procedures with relatively low risks of readmission.
As CMS announced in 2012 that there would be penalties for facilities with excessive readmissions for patients with congestive heart failure, pneumonia, and myocardial infarction, there has been an increased interest in determining overall readmission rates and predictors of unplanned hospital stays.16 Although there are not yet any penalties or defined outcome measures specific to spinal surgery, there are efforts to better define clinically relevant readmission rates, and in particular, the factors that influence these rates.
Our study adds to the literature in several ways. Our study adds to previous hospital-wide data by identifying the facility-level predictors of readmission.8 As all facilities have subtly different case mixes, it is important to establish a benchmark range. Our study also differed from previous studies that investigated only several CPT codes (ie, cervical fusions9 and lumbar decompressions12,15) with the national NSQIP database.9, 14, 15 As a result, our study sample size was smaller than NSQIP investigations, but it better reflects the variance in types of procedures at a typical hospital. Particularly, if hospitals are to be evaluated on hospital-wide readmissions, it is critical to have a relevant hospital-wide readmission rate. Also, these NSQIP investigations do not address procedures involving the thoracic spine or complex spinal deformities, which are relatively common procedures at tertiary institutions.
NSQIP uses clinical reviewers to sample a certain amount of consecutive cases every 8 days depending on the size of the hospital. Although this process tries to optimize case selection and case mix, by not following consecutive patients, NSQIP can miss rare pathologies, which can affect the measurement of overall hospital-wide readmission rates. In addition, by creating our own database and following all patients that had spine surgery in 1 calendar year, we avoided some of the disadvantages of using the NSQIP data. By using our own database, we were able to determine the reason for the readmission and if it was procedure related. Finally, we were able to assess some variables not captured by NSQIP such as whether a procedure was MIS or Open.
Upon review by attending neurosurgeons utilizing previously cited criteria, less than half of our readmissions were deemed procedure related, and there was significant agreement between the reviewers. Although this seems like quite a significant divergence from what would be reported as the all-cause readmission rate, the proportion of readmissions deemed procedure-related readmissions is actually slightly higher than that in a previous study.12 Of these procedure-related admissions, most were due to postprocedure complications or surgical site infections. Procedure-unrelated readmissions were primarily new community-acquired conditions such as nonrelated infection and trauma, but progression of underlying disease and exacerbation of pre-existing chronic conditions were also contributors. Thus, our study provides further evidence that calculating readmission rates from administrative datasets, which can only identify whether or not a patient was readmitted, do not provide an accurate representation of the clinical picture. Thus, manual review or improved codes with which to classify readmissions are needed if readmission rates are to be used to compare institutions or modify payment.
The 3 variables that significantly impacted risk of readmission (number of spinal levels performed during the surgery, number of comorbidities present in the patient at the time of surgery, and whether the admission required an ICU stay) give guidance on where to target improvement efforts. The fact that patients with multiple comorbidities are readmitted more often supports continued efforts to optimize medical management before a patient is taken to the operating room. Implementing algorithms such as those developed by Sugrue et al17 have been shown to reduce readmissions in these high-risk patients. Similarly, preoperative counseling should be given to high-risk patients to reduce the likelihood of an ICU stay.
Limitations to this study include a smaller sample size than similar studies; however, as discussed previously, our experimental design gave us the ability to extract more detailed information than those studies with larger numbers. In addition, as this is a retrospective study, our information was limited to the information that was coded in the electronic medical record at the time of the surgery, which may have decreased our ability to detect selection bias. Also, past studies have demonstrated that social factors such as income level and family support have strong effects on readmissions, and although we saw no association between insurance type and risk of readmission,18 we were unable to assess income level and family support. In our study, we were only able to track patients who were readmitted to our hospital, but not if they were readmitted to an outside institution. As Northwestern Memorial Hospital is a tertiary referral institution, it is possible that several patients may have presented for readmission at hospitals closer to their home, and thus been missed in our analysis, leading to an artificially low readmission rate. However, 94.3% of patients were seen at follow-up clinic appointments after 30 days, so this is unlikely to significantly bias our conclusions. Finally, as mentioned previously, our study had both advantages and disadvantages by being a single-center study. Our hospital is a large academic institution, which had 19 different surgeons operating during the study period on a wide variety of cases and included both neurosurgeons and orthopedic surgeons who completed spine fellowships, as well as neurosurgeons and orthopedic surgeons who did not complete spine fellowships. Thus, our results generalize best to academic centers and large spine practices.
Additional studies that could build on this work include determining the financial impact of these readmissions and analyzing predictors of readmission within 2 years of follow-up. Strategies to minimize the impact of the identified predictors should also be developed.
Our study shows that, after multivariate analysis, an association exists between the rate of readmission within 30 days following spinal procedures and the number of spinal levels performed during the surgery, the number of patient comorbidities present at the time of surgery, and whether the admission required an ICU stay. Our study also further supports the contention that administrative readmission rates are inaccurate in determining the clinically relevant readmission rate after spinal surgery. Efforts toward modifying medical risk factors and comorbidities may lead to decreased 30-day readmission rates.
- Records from 1187 consecutive spinal surgeries at a single institution were retrospectively reviewed and data were collected that described the patient, surgical procedure, hospital course, complications, and readmissions.
- The overall readmission rate was 6.1%.
- Of the readmissions, 37.5% of readmissions were deemed procedure related upon attending review, leading to a procedure-related readmission rate of 2.3%.
- Upon multivariate analysis, only 3 variables were found to be significant predictors of readmission: 2 or more patient comorbidities, an admission to the ICU, and each additional spinal level involved.
1. Porter ME. What is value in health care? N Engl J Med
2. Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med
3. Goodman DC, Fisher ES, Chang C. Robert Wood Johnson Foundation.
4. Medicare Hospital Compare Quality of Care
5. One Hundred
Eleventh Congress of the United States of America. HR 3590, Sec 3025: Hospital Readmissions
Reduction Program; 2010.
6. Rau J. ArmedWith Bigger Fines, Medicare To Punish 2225 Hospitals For Excess Readmissions
. Kaiser Health News [serial online]. August 02, 2013. Accessed.
7. Weinstein JN, Lurie JD, Olson PR, et al. United States’ trends and regional variations in lumbar spine surgery
: 1992-2003. Spine
8. Amin BY, Tu TH, Schairer WW, et al. Pitfalls of calculating hospital readmission rates based on nonvalidated administrative data sets: presented at the 2012 Joint Spine Section Meeting: clinical article. J Neurosurg Spine
9. Lovecchio F, Hsu WK, Smith TR, et al. Predictors of thirty-day readmission after anterior cervical fusion
10. CMS. Acute Inpatient PPS 2014
. Services CfMM ed. cms.gov.
11. 2002; Fessler RG. 51:Siii–Siii10.
12. Shah MN, Stoev IT, Sanford DE, et al. Are readmission rates on a neurosurgical service indicators of quality
of care? J Neurosurg
13. Wang MC, Shivakoti M, Sparapani RA, et al. Thirty-day readmissions
after elective spine surgery
for degenerative conditions among US Medicare beneficiaries. Spine J
14. Pugely AJ, Martin CT, Gao Y, et al. Causes and risk factors for 30-day unplanned readmissions
after lumbar spine surgery
15. Kim BD, Smith TR, Lim S, et al. Predictors of unplanned readmission in patients undergoing lumbar
decompression: multi-institutional analysis of 7016 patients. J Neurosurg Spine
16. Sacks GD, Dawes AJ, Russell MM, et al. Evaluation of hospital readmissions
in surgical patients: do administrative data tell the real story? JAMA Surg
17. Sugrue PA, Halpin RJ, Koski TR. Treatment algorithms and protocol practice in high-risk spine surgery
. Neurosurg Clin N Am
18. Joynt KE, Orav EJ, Jha AK. Thirty-day readmission rates for Medicare beneficiaries by race and site of care. JAMA