INTRODUCTION
Lumbar spinal stenosis (LSS) is a common cause of low back pain and associated disability [1]. Although most patients with LSS are treated in primary care, not all receive satisfactory treatment responses, which can lead to hospitalization and spine surgery [2]. In the United States (US), hospitalizations for LSS increased by 19.9% between 2016 and 2019 [3]. Readmissions after spine surgery for LSS have also remained high; reducing these readmissions can improve healthcare quality and decrease healthcare costs [4]. Although several studies have previously assessed readmissions after spine surgery for patients with LSS, these studies have had key limitations, namely: single-site inclusions [5,6], restrictions to a single payer system [7-9], and data being now out of date [7-10]. Some studies have used nationally representative data to explore readmissions after elective lumbar spine surgery [11-13]. However, these studies combined multiple spinal diseases (e.g., one study included lumbar disc herniation, lumber stenosis, acquired spondylolisthesis, and degeneration of lumbar or lumbosacral intervertebral disc [11]), which may bias the results due to the population heterogenicity [14]. Additionally, patients with LSS are usually not young, which may require an extended recovery period [15]. Previous relevant studies often assessed 30- or 90-day readmissions, but monitoring longer-term outcomes following surgery is required to better understand surgical quality and safety [16]. This need for extended monitoring was endorsed by the National Institutes of Health and the American College of Surgeons in 2016 [17]. Moreover, studies in other areas have emphasized that the causes and predictive factors of readmissions may vary across different periods after surgery [18-20]. This information is useful for clinicians and policymakers in their efforts to reduce and manage readmissions, as different strategies may be needed at different periods after surgery.
Therefore, in this nationally representative retrospective cohort study of patients with LSS who underwent spine surgery, we determined the rates, causes, and predictive factors of readmissions at different periods following spine surgery, up to 180 days.
MATERIALS AND METHODS
The study was reported following the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) reporting guideline [21]. As this was a secondary analysis of publicly available deidentified databases, Institutional Review Board approval and patient written informed consent were not required as it did not involve human participants.
1. Data Sources
The data used in this study came from the Nationwide Readmissions Database (NRD), Healthcare Cost and Utilization Project (HCUP), Agency for Healthcare Research and Quality [22]. It provides nationally representative data on hospital readmissions in the US. For example, the 2019 NRD data account for 61.8% of the total US resident population and 60.4% of all US hospitalizations [23]. When deciding which data waves to include, 2 types of issues were considered. First, the diagnosis and procedure groups file, available in the NRD since 2018, is required because the Clinical Classifications Software Refined (CCSR) for the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnoses are needed to determine the causes of readmissions [24,25]. Second, the coronavirus disease 2019 lockdown significantly impacted health service use, potentially providing biased estimates if data from the pandemic period was included [26]. Therefore, this study included adults over 18 years old across two annual waves (2018 and 2019).
2. Outcomes
Primary outcomes included the rates, causes, and predictive factors of first readmissions following spine surgery (identified through ICD-10 Procedure Coding System [ICD-10-PCS]; detailed in Supplementary Table 1) in patients with LSS (identified through ICD-10-CM; detailed in Supplementary Tables 2 and 3) [19]. Based on the discharge date of initial hospitalization, 4 periods were included: 0 and 7 days, 8 and 30 days, 31 and 90 days, and 91 and 180 days after the initial hospitalization [19]. Patients who died during the initial hospitalization were excluded as the proportion of death is small (detailed in Fig. 1). To ensure accurate analysis of the NRD dataset, it is important to consider the issue of immortal time bias [27]. For instance, patients admitted in December may not have sufficient follow-up time to observe a 30-day follow-up. To mitigate this potential bias, we have employed the following strategy: (1) exclusion of patients admitted in December for the analysis of 7 and 30 days; (2) exclusion of patients admitted from October to December for the analysis of 90 days; (3) exclusion of patients admitted from July to December for the analysis of 180 days [19].
Causes of readmissions were defined based on the principal diagnosis category (grouped through CCSR for ICD-10-CM diagnoses; detailed in Supplementary Tables 4 and 5) [28]. The CCSR for ICD-10-CM diagnoses aggregates more than 70,000 ICD-10-CM diagnosis codes into over 530 clinically meaningful categories. Version 2024.1 was used and can handle ICD-10-CM codes from October 2015 through September 2024.
Potential predictive factors were identified and confirmed based on prior literature (detailed in the Supplementary Methods) and their identifiability in the NRD. A total of 24 predictive factors (detailed in Supplementary Methods) were explored: age, sex, insurance type, median household income, patient location, day of admission, length of hospital stay, discharge status, admission type, Elixhauser Comorbidity Index [29], type of surgery, depression, anxiety, rheumatoid arthritis, osteoarthritis, osteoporosis, overweight, obesity, hypertriglyceridemia, hypertension, alcohol use disorders, drug use disorders, smoking, and sleep disorders.
3. Statistical Analysis
For each period, rates and causes of readmissions were reported as proportions; multivariable logistic regression was used to examine the association (reported as odds ratio [OR] with 95% confidence interval [CI]) between potential predictive factors and the readmission, with adjustments for all variables. The cumulative readmission rates (7 days, 30 days, 90 days, and 180 days) and subgroup (based on those mentioned above potential predictive factors) results were also estimated to facilitate communication. To ensure causes of readmissions meet the interests of clinicians and policymakers, a three-level hierarchy was used to represent them: level one, surgical site-related and nonsurgical site-related causes; level two, categories through CCSR for ICD-10-CM diagnoses; level three, conditions through ICD-10-CM coding system.
As the NRD collected data through the complex sampling strategy, weights were used to ensure that the estimates were nationally representative, and weights and design variables were included to obtain unbiased estimates and standard errors [30]. Complete case analysis was performed as the proportion of missing data was small (detailed in Supplementary Table 6). Based on the requirements for publishing with HCUP data, cell sizes less than or equal to 10 were not reported. The significance was defined as 2-sided p-value at 0.05. Data were analyzed through Stata version 17.0 (StataCorp LLC, College Station, TX, USA) and IBM SPSS Statistics ver. 27.0 (IBM Co., Armonk, NY, USA).
RESULTS
The analysis sample included 180,281 patients (Fig. 1). Characteristics of included patients are presented in Table 1. The study participants had a mean age of 65.4 years. 51.7% of the participants were female, 85.6% had at least one comorbidity, and 78.8% underwent fusion.
1. Readmissions Between 0 and 7 Days
Of 2.4% of the participants readmitted, the top ten causes (Table 2) were: other surgical care complication (e.g., infection) (10.7%), postoperative nervous system complication (e.g., hematoma and seroma of a nervous system structure) (7.9%), septicemia (6.8%), nervous system pain and pain syndromes (6.0%), spondylopathies/spondyloarthropathy (e.g., spinal stenosis) (5.8%), postoperative musculoskeletal system complication (e.g., hematoma and seroma of a musculoskeletal structure) (4.4%), other nervous system disorders (3.3%), acute pulmonary embolism (2.6%), urinary tract infections (2.5%), and intestinal obstruction and ileus (2.3%). Among the ten causes listed above, the proportion of surgical site-related complications is higher than that of nonsurgical site-related complications (28.9% vs. 23.3%, p<0.001) (Supplementary Table 7). At level 3, the top 3 conditions were infection (8.3%), acute postsurgical pain (5.9%), and hematoma and seroma of a nervous system structure (4.5%).
Estimates for statistically significant predictive factors are provided in Table 3, while estimates for other predictors are detailed in Supplementary Table 8. Several predictive factors were identified as being associated with an increased risk of readmissions. For example, patients aged 85 years or older had a higher risk compared to those aged 18 to 44 years (OR, 1.59; 95% CI, 1.18 to 2.15; p=0.003). Several predictive factors were identified as being associated with a reduced risk of readmissions. For example, being female compared to male was associated with a reduced risk (OR, 0.78; 95% CI, 0.72–0.84; p<0.001).
2. Readmissions Between 8 and 30 Days
Of 3.5% of the participants readmitted, the top 10 causes (Table 4) were: other surgical care complication (e.g., infection) (29.2%), spondylopathies/spondyloarthropathy (e.g., spinal stenosis) (7.7%), septicemia (6.2%), internal orthopedic device or implant complication (4.5%), postoperative nervous system complication (e.g., hematoma and seroma of a nervous system structure) (4.2%), postoperative musculoskeletal system complication (e.g., hematoma and seroma of a musculoskeletal structure) (3.4%), acute pulmonary embolism (3.4%), acute renal failure (1.9%), urinary tract infections (1.9%), and other nervous system disorders (e.g., toxic encephalopathy) (1.7%). Among the 10 causes listed above, the proportion of surgical site-related complications is higher than that of nonsurgical site-related complications (41.4% vs. 22.7%, p<0.001) (Supplementary Table 7). At level 3, the top 3 conditions were infection (22.3%), disruption of wound (5.4%), and sepsis (4.3%).
Estimates for statistically significant predictive factors are provided in Table 3, while estimates for other predictors are detailed in Supplementary Table 8. Several predictive factors were identified as being associated with an increased risk of readmissions. For example, discharge to home health care compared to routine discharge was associated with an increased risk (OR, 1.24; 95% CI, 1.14–1.35; p<0.001). Several predictive factors were identified as being associated with a reduced risk of readmissions. For example, having private insurance compared to Medicare was associated with a reduced risk (OR, 0.72; 95% CI, 0.65–0.80; p<0.001).
3. Readmissions Between 31 and 90 Days
Of 3.7% of the participants readmitted, the top ten causes (Table 5) were: spondylopathies/spondyloarthropathy (e.g., spinal stenosis) (12.6%), other surgical care complication (e.g., injection) (11.4%), internal orthopedic device or implant complication (6.3%), osteoarthritis (6.0%), septicemia (5.3%), postoperative musculoskeletal system complication (e.g., hematoma and seroma of a musculoskeletal structure) (2.0%), heart failure (1.9%), pneumonia (1.9%), postoperative nervous system complication (e.g., hematoma and seroma of a nervous system structure) (1.8%), and urinary tract infections (1.7%). Among the ten causes listed above, the proportion of nonsurgical site-related complications is higher than that of surgical site-related complications (29.3% vs. 21.4%, p<0.001) (Supplementary Table 7). At level three, the top three conditions were infection (8.2%), spinal stenosis (6.0%), and sepsis (3.5%).
Estimates for statistically significant predictive factors are provided in Table 6, while estimates for other predictors are detailed in Supplementary Table 9. Several predictive factors were identified as being associated with an increased risk of readmissions. For example, patients aged 85 years or older had a higher risk compared to those aged 18 to 44 years (OR, 1.42; 95% CI, 1.09–1.84; p=0.009). Several predictive factors were identified as being associated with a reduced risk of readmissions. For example, having private insurance compared to Medicare was associated with a reduced risk (OR, 0.79; 95% CI, 0.71–0.89; p<0.001).
4. Readmissions Between 91 and 180 Days
Of 4.3% of the participants readmitted, the top 10 causes (Table 7) were: spondylopathies/spondyloarthropathy (e.g., spinal stenosis) (17.5%), osteoarthritis (16.7%), septicemia (4.7%), internal orthopedic device or implant complication (5.4%), postoperative musculoskeletal system complication (e.g., hematoma and seroma of a musculoskeletal structure) (2.5%), heart failure (2.1%), cardiac dysrhythmias (1.8%), other surgical care complication (e.g., infection) (1.7%), coronary atherosclerosis and other heart disease (1.7%), and pneumonia (1.5%). Among the ten causes listed above, the proportion of nonsurgical site-related complications is higher than that of surgical site-related complications (46.04% vs. 9.72%, p<0.001) (Supplementary Table 7). At level 3, the top 3 conditions were unilateral primary osteoarthritis of hip (9.2%), spinal stenosis (8.7%), and unilateral primary osteoarthritis of knee (4.9%).
Estimates for statistically significant predictive factors are provided in Table 6, while estimates for other predictors are detailed in Supplementary Table 9. Several predictive factors were identified as being associated with an increased risk of readmissions. For example, patients aged 85 years or older had a higher risk compared to those aged 18 to 44 years (OR, 1.50; 95% CI, 1.11–2.03; p=0.008). Several predictive factors were identified as being associated with a reduced risk of readmissions. For example, having private insurance compared to Medicare was associated with a reduced risk (OR, 0.78; 95% CI, 0.70–0.87; p<0.001).
5. Cumulative Readmission Rates
The cumulative readmission rates at 7, 30, 90, and 180 days were 2.4%, 5.9%, 9.3%, and 12.1% respectively (detailed in Supplementary Table 10). These rates varied a lot across different subgroups. For example, the lowest rate was 7.6% in patients with an Elixhauser Comorbidity Index score of 0, while the highest rate was 25.7% in patients transferred to a short-term hospital.
DISCUSSION
1. Principal Findings
This large nationally representative retrospective cohort study showed that about 6% of patients with LSS who underwent spine surgery were readmitted within 30 days and 12% by 180 days. The causes of readmissions varied across different periods: surgical site-related causes predominated within the first 30 days, whereas nonsurgical site-related causes were more prevalent from 31 to 180 days. Considering CCSR for ICD-10-CM diagnoses, other surgical care complication (e.g., infection) was the most prevalent cause within the first 30 days, whereas spondylopathies/spondyloarthropathy (e.g., spinal stenosis) became the most prevalent cause from 31 to 180 days. Considering conditions directly identified by the ICD-10-CM coding system, infection was the most prevalent cause within the first 90 days, whereas unilateral primary osteoarthritis of hip became the most prevalent cause from 91 to 180 days. The predictive factors associated with readmissions also varied across different periods. For example, patients who underwent fusion was associated with a decreased risk of readmissions between 31 and 180 days, rather than between 0 and 30 days.
2. Comparison With Previous Studies
We systematically searched (detailed in Supplementary Methods) for studies focused on readmissions in US patients with LSS who underwent spine surgery and identified five relevant studies. Therefore, we compared the rates of readmissions, causes of readmissions, and predictive factors of readmissions from our study with these 5 studies [5,7-10].
For estimating rates of readmissions, three studies reported 30-day readmission rates, with estimates of 9.1%, 3.7%, and 4.0% [5,7,10]. One study reported a 90-day readmission rate, with an estimate of 7.2% [5]. One study reported a 1-year readmission rate, with an estimate of 9.7% in patients undergoing fusion with decompression and 7.2% in patients undergoing decompression alone [8]. Another study reported a 1-year readmission rate with an estimate of 17.5% [9]. These differences, either within different studies or compared with ours, could be due to the following explanations (detailed in Supplement Methods): firstly, variations in inclusion and exclusion criteria. For example, while Deyo et al. [7] and Modhia et al. [8] broadly categorized surgical procedures as “decompression” or “fusion” without providing detailed coding specifications, and Basques et al. [10] relied on ICD-9-CM codes, our study used precise ICD-10-PCS codes to identify specific procedures such as “discectomy,” “diskectomy,” “lamilaminectomy,” and “laminotomy”; secondly, differences in the representativeness of the population. Prior studies, such as those by Deyo et al. [7] and Ong et al. [9], were limited to Medicare beneficiaries aged 65 and older, while Ilyas et al. [5] focused on a single-center cohort. In contrast, our study utilized the NRD database, which provides nationally representative data on hospital readmissions in the US. And thirdly, the timeframes for recruiting patients. Early studies, such as Ilyas et al. [5], included data from 2014–2015 and focused solely on readmissions within 90 days after surgery, without evaluating long-term outcomes. Our study used nationally representative data and adhered to strictly defined inclusion and exclusion criteria. It provided estimates at different periods after surgery, facilitating further comparison with estimates from other countries. The persistent rise in readmission rates indicates that quality improvement efforts to reduce the burden of readmissions should continue for at least 180 days after spine surgery.
For analyzing causes of readmissions, the study by Ilyas et al. [5] reported a binary cause (i.e., surgical and nonsurgical) for 90-day readmissions, while the study by Basques et al. [10] reported suspected reasons for 30-day readmissions. Our study extended previous studies through 2 aspects. Firstly, our study used a systematic method to define causes. Secondly, our study reported these causes at different periods after surgery, allowing us to observe relevant changes over time. Contrary to expectations, infections peaked between 8 and 30 days rather than in the initial 7 days, which may indicate insufficient management of postoperative infections. Although current guidelines acknowledge the need for managing postoperative infections, they do not provide specific strategies [31,32]. Thus, one feasible approach is to educate patients on the signs and symptoms of surgical site infections prior to hospital discharge, enabling them to seek appropriate and timely care at home and potentially preventing some readmissions. The reasons why infections are most common between 8 and 30 days after surgery remain unclear. One potential direction for future studies is to conduct genomic-based microbial tracking, starting from the built environment (i.e., the operating room), through the patient’s own microbiome, and ultimately to the surgical site [33]. This approach could help elucidate the true pathogenesis of infections and enable the development of more targeted interventions. A potential reason why unilateral primary osteoarthritis of the hip became the most prevalent cause between 91 and 180 days could be related to the hip-spine syndrome [34]. This condition suggests that the pathological changes caused by degenerative diseases in both the hip and the spine may be interrelated. Although the exact reason should be further explored, surgeons should communicate this type of readmission to patients.
For analyzing predictive factors of readmissions, the previous 5 studies each included at least one factor, but none of them comprehensively explored this area [5,7-10]. Our study identified potential predictive factors by systematically reviewing prior literature. The results indicated that the role of some predictors was inconsistent across different periods. For example, patients with alcohol use disorders were associated with a higher risk of readmission after 90 days after surgery. Although we could not identify the reason in the NRD dataset, future studies should assess whether certain changes, such as behavioral modifications, occurred in these patients at specific time points after surgery to improve management in specific subpopulations. We broadened the definition of comorbidity by using a systematic approach rather than selecting a few specific types of conditions, and the consistent results across different periods showed that 2 or more comorbidities were associated with increased readmissions, and the association became stronger as the number of comorbidities increased. This information highlights the need for ongoing management of patients with a high number of comorbidities. We also assessed the role of the type of surgery and the results appear to contradict traditional thinking as fusion being associated with a higher risk of readmission compared to decompression alone. A previous study found that patients who underwent fusion for LSS had a higher 1-year readmission rate compared to those who had decompression alone (9.7% vs. 7.2%, p=0.03) [8]. One potential reason could be spinal instability. A previous randomized controlled trial by Ghogawala et al. [35] reported the cumulative risk of reoperation in both the decompression-only and fusion groups. The findings indicated that patients in the decompression-only group had a higher risk of early reoperation, beginning as early as 4 months post-surgery, due to instability. Surgeons should effectively communicate this information to patients to help guide the selection of the most appropriate surgical approach.
3. Limitations
Several limitations should be mentioned. Firstly, the NRD could not track patients who were readmitted to another state, so the readmission rates may be underestimated [23]. Secondly, the NRD is an administrative database without data linkage to other databases. Therefore, variables requiring additional tests, such as genetic factors, those needing patient responses, such as quality of life, or data from other healthcare settings, such as prescriptions from general practice, could not be included in this study. Some of these variables are potential predictive factors (detailed in Supplementary Methods), which should be explored further. Thirdly, some lifestyle factors were defined through the ICD-10-CM coding system, which may bring biases. For example, using a self-reported survey, smoking status can be categorized as nonsmoker, former smoker, or current smoker, with the current smoker group further classified by varying levels of smoking intensity. This approach provides more comprehensive information compared to the binary classification available through the ICD-10-CM coding system.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Full text via PMC
Download Citation
Supplement
Print
