Use of the LACE ++ index to predict readmissions after single-level lumbar fusion

OBJECTIVE Spinal fusion is one of the most common neurosurgical procedures. The LACE (length of stay, acuity of admission, Charlson Comorbidity Index [CCI] score, and emergency department [ED] visits within the previous 6 months) index was developed to predict readmission but has not been tested in a large, homogeneous spinal fusion population. The present study evaluated use of the LACE + score for outcome prediction after lumbar fusion. METHODS LACE + scores were calculated for all patients (n = 1598) with complete information who underwent single-level, posterior-only lumbar fusion at a single university medical system. Logistic regression was performed to assess the ability of the LACE + score as a continuous variable to predict hospital readmissions within 30 days (30D), 30–90 days (30–90D), and 90 days (90D) of the index operation. Secondary outcome measures included ED visits and reoperations. Subsequently, patients with LACE + scores in the bottom decile were exact matched to the patients with scores in the top 4 deciles to control for sociodemographic and procedural variables. RESULTS Among all patients, increased LACE + score significantly predicted higher rates of readmissions in the 30D (p < 0.001), 30–90D (p = 0.001), and 90D (p < 0.001) postoperative windows. LACE + score also predicted risk of ED visits at all 3 time points and reoperations at 30–90D and 90D. When patients with LACE + scores in the bottom decile were compared with patients with scores in the top 4 deciles, higher LACE + score predicted higher risk of readmissions at 30D (p = 0.009) and 90D (p = 0.005). No significant difference in hospital readmissions was observed between the exact-matched cohorts. CONCLUSIONS The present results suggest that the LACE + score demonstrates utility in predicting readmissions within 30 and 90 days after single-level lumbar fusion. Future research is warranted that utilizes the LACE + index to identify strategies to support high-risk patients in a prospective population.

to predict 30-day readmission after hospital discharge. 6he original developers of the LACE index have since updated it to the LACE+ index, which includes age, sex, and history of hospital admissions in its calculation. 7We have previously demonstrated that this metric can be easily automated for use with electronic health record (EHR) systems and therefore represents an enticing risk stratification tool for surgical populations. 8lthough a handful of studies have utilized the LACE+ index to predict adverse postoperative events across an array of subspecialties, [9][10][11][12] including neurosurgery, 8,[13][14][15] this tool has not been tested in a large, homogeneous spinal fusion population.Spinal fusion is one of the most common neurosurgical procedures and bears an appreciable risk of postoperative complications. 16Therefore, the present study aimed to address this gap and to evaluate the efficacy of using the LACE+ score to predict postoperative outcomes of patients undergoing single-level, posterior-only lumbar fusion.Moreover, coarsened exact matching (CEM) was employed to control for confounding sociodemographic and procedural variables-such as sex, race, and direct surgical cost-and to isolate the ability of the LACE+ score to predict patient outcomes.

Ethics Committee Approval
This study was approved by the IRB of the Hospital of the University of Pennsylvania.All ethical guidelines and rules were followed to protect patient privacy.

Sample Selection and Data Collection
Consecutive adult patients (n = 4333) who underwent single-level lumbar fusion at a single, multihospital, 1659-bed academic medical center from June 7, 2013, to April 29, 2019, were retrospectively enrolled (Fig. 1).Subsequently, patients who underwent anterior-posterior lumbar fusion and nonroutine operations were excluded from analysis.Additionally, patients with incomplete data, and therefore a LACE+ score could not be calculated and CEM could not be performed, were removed.The remaining patients (n = 1598) composed the sample population.
Records with complete information (sex, admission type, length of stay, CCI score, recent ED visits, surgical history, history of hospital admission) were utilized to calculate the composite LACE+ score (range −1 to 90) for each patient (Table 1 and Supplemental Table 1).Additional patient information (delineated below) was utilized to control for confounding variables in CEM.The primary outcome variable, unplanned hospital readmission, was recorded within the postsurgical windows of 30 days (30D), 30-90 days (30-90D), and 90 days (90D).Addi-  tional adverse events included ED visits and reoperations within 30D, 30-90D, and 90D after the index operation.
We reported the number of patients who experienced an adverse event during the designated postsurgical time window.Patient data were captured with the EpiLog tool, a nonproprietary data acquisition system created for quality improvement initiatives. 17The IRB granted a waiver for informed consent because this study was considered to have minimal risk to patients.

Matching
CEM was performed after regression analysis.Patients with LACE+ scores in the bottom decile were matched with those patients with scores in the top 4 deciles to control for confounding variables.These cutoffs, which were previously used in other studies, 18 were used to evaluate the extremes of LACE+ scores and achieve adequate statistical power with a large population of patients available for exact matching.Unlike propensity score matching, which converts numerous covariates into a single composite value for matching, CEM utilizes raw covariate values for 1:1 exact matching. 19With CEM, an exact match refers to a match for every matching variable, and unmatched patients are subsequently removed from the analysis.Therefore, CEM results in fewer patients included in the analysis but offers superior control of confounding variables.
The present study generated matches for 9 unique patient and procedural characteristics, including median household income, direct surgical cost (total cost of surgical supplies and implants), race, insurance status, previous surgical history, CCI score, BMI, sex, and tobacco use.Matching was performed in a binary manner for median household income and direct surgical cost (either above or below the median value of the population), race (White or non-White), insurance status (private or other), and previous surgical history (presence or absence of operations within 30D preceding the index operation).BMI and CCI score were matched in a tertiary manner, with all individuals stratified into either a low, medium, or high cohort.Sex and tobacco use were exactly matched.

Statistical Analysis
Univariate logistic regression analysis of the sample population was performed to evaluate the capability of a single-point increase in LACE+ score to predict adverse postoperative events.Next, risk of adverse postsurgical events was compared between the patients with LACE+ scores in the lowest quartile (range 8-21; n = 403 [25.22%]) and those with scores in the highest quartile (range 43-81; n = 387 [24.22%]), and then risk was compared between those with scores in the bottom decile (range 8-17; n = 160 [10.01%]) and those with scores in the top 4 deciles (range 31-81; n = 672 [42.05%]).Finally, CEM was performed to match the patients with LACE+ scores in the bottom decile and those with scores in the top 4 deciles.The Pearson chi-square test was employed to compare categorical demographic characteristics between the cohorts before and after CEM, and the Wilcoxon rank-sum test was used to compare continuous variables.The McNemar test was performed to compare risk of postsurgical outcomes.
Matches were generated with the MatchIt programming package in R Statistics (R Core Team).All endpoint analyses utilized SAS version 9.4 (SAS Institute Inc.), with significance set at p < 0.05.

Patient Characteristics of the Sample Population
Among all patients included in the analysis (n = 1598), 46.31% of patients were male and 79.46% were White (Table 2).The mean age was 61.70 years.The LACE+ scores of the population ranged from 8 to 81.

Patient Characteristics of the Exact-Matched Cohorts
After CEM, 111 patients were included in each cohort (Table 3).Patients with LACE+ scores in the bottom decile did not demonstrate significant differences in sex, race, tobacco use, duration of follow-up, or total cost of surgery.Patients with higher LACE+ scores were significantly older, had higher CCI scores, and experienced longer length of stay (all of which are components of the LACE+ index).
When we compared the patients with LACE+ scores in the lowest quartile with those patients with scores in the highest quartile, higher LACE+ score predicted higher risk of readmissions at 30D (p = 0.003), 30-90D (p = 0.016), and 90D (p < 0.001) after index surgery (Fig. 2B and Table 4).In addition, LACE+ score predicted postoperative risk of ED visits at 90D but not 30D or 30-90D.LACE+ score also predicted risk of reoperation at 30-90D but not 30D or 90D.
When we compared the patients with LACE+ scores in the bottom decile with those patients with scores in the top 4 deciles, higher LACE+ score predicted higher risk of readmissions at 30D (p = 0.009) and 90D (p = 0.005) but not 30-90D after index surgery (Fig. 2C and Table 4).LACE+ score did not predict postoperative risk of ED visits or reoperations.

Outcomes of the Matched Cohorts
There were no significant differences in the risks of readmissions after index surgery between the matched cohorts (Fig. 3 and Table 5).Additionally, no significant differences in risks of ED visits and reoperations were observed between cohorts.

Discussion
The LACE+ index predicted risks of short-term and midterm unplanned hospital readmissions after single-level, posterior-only lumbar fusion.Moreover, LACE+ score was associated with risks of additional postoperative complications, including ED visits and repeat neurosurgical intervention.Thus, the LACE+ index appears to have utility for the prediction of postoperative outcomes in this population.
In the present study, univariate regression analysis was employed to evaluate the effect of single-point increase in LACE+ score on the risks of readmissions and other adverse postsurgical events.Although this methodology allowed for a robust analysis of LACE+ score as a continuous variable in these patients, it did not account for other factors that may predispose patients to inferior outcomes.As such, the present study utilized CEM to control for confounding variables-including race [20][21][22] and smoking status 23,24 -known to independently affect postsurgical outcomes after elective spinal surgery.Of note, insurance status, CCI score, and prior surgical history were significantly different between the exact-matched cohorts.These findings are not entirely unexpected.Because age is incorporated into the LACE+ score, it is possible that patients with higher LACE+ scores due to older age may be more likely to have Medicare insurance.Also, as a result of older age, it is plausible that these patients have undergone more surgical procedures.Meanwhile, CCI score is a direct component of the LACE+ index.
]13,14   For the present analysis, we utilized a decile-based matching method that included an increased number of patients with higher risk of readmission (as reflected by the higher LACE+ scores).In doing so, this analysis examined the extremes of LACE+ scores while ensuring a large sample of patients who underwent exact matching.
In the present study, univariate regression analysis demonstrated a significant positive correlation between LACE+ score and adverse postsurgical outcomes.The ORs of the entire population analysis (range 1.02-1.04for statistically significant values) appeared weaker than those of the quartile-based (range 1.71-3.06)and decile-based (range 3.42-3.94)analyses.That is, although a single-point increase in LACE+ score was associated with increased risk of adverse outcomes by 2%-4%, patients with LACE+ scores in an upper quartile or decile may have increased odds for these events by as much as 294%.This finding was not unexpected: the LACE+ index has a large range, so single-unit increases in LACE+ score would understandably result in smaller effects than large differences (e.g., between low and high deciles).No significant differences were observed between the exact-matched cohorts, although the results trended toward higher LACE+ scores being predictive of inferior outcomes.These findings suggest that the variables incorporated into the LACE+ index are useful for outcome prediction across a large population.The advantage of the LACE+ index is that it condenses multiple crucial patient characteristics into an easy-tointerpret score, which can be easily integrated into EHR systems.Nevertheless, additional research with a larger, multiinstitutional population is warranted to further characterize the utility of the LACE+ index for lumbar fusion populations.

TABLE 4. Prematch results of logistic regression analysis used to examine the association of LACE+ score with outcomes after singlelevel, posterior-only lumbar fusion for all patients included in this analysis (n = 1598), patients with LACE+ scores in quartile 1 (n = 403) compared with those patients with scores in quartile 4 (n = 387), and patients with LACE+ scores in the bottom decile (n = 160) compared with those patients with scores in the top 4 deciles (n = 672)
Use of the LACE+ index, in the context of elective lumbar fusion surgery, may help guide management efforts in the most resource-efficient manner.Of note, multiple components of the LACE+ index are not readily modifiable (e.g., comorbidity status, previous ED visits), and some components cannot be calculated until discharge (length of stay).However, with integration into institutional EHR systems, the LACE+ index may have the potential to revolutionize discharge planning by providing computed scores in real time.Postoperative resources could be channeled toward patients at high risk for adverse postsurgical events, as predicted by the LACE+ score.For example, patients with a high LACE+ score at the time of discharge may be offered additional postoperative support, such as in-home nursing services.
Although the present study did not aim to characterize the precise reasons underlying unplanned readmissions in this population, other studies have pointed toward inadequate pain management, surgical site infections, and thromboembolic phenomena as the most common causes. 25As such, specific strategies, such as improved outreach tailored toward these events, may benefit highrisk patients, as identified with a high LACE+ score.
The goal of the present study was to evaluate the utility of the LACE+ index for stratification of patients according  to risk of adverse events after posterior-approach spinal fusion, irrespective of surgical indications or procedural techniques.This approach was employed to generate information regarding the large-scale predictive capabilities of this index, which could be subsequently utilized to inform risk mitigation efforts at the institutional scale.From a system-wide standpoint, risk stratification tools such as the LACE+ index have the potential to transform healthcare delivery and resource allocation.With increased attention turned toward the minimization of avoidable healthcare expenditures, the incidence and prevention of adverse events such as unplanned readmission are likely to remain a priority among healthcare providers.Therefore, easily accessible data-driven tools that give a snapshot of a patient's risk profile represent a tantalizing opportunity for healthcare innovation.

Limitations
The present study is not without its limitations.The retrospective analysis was performed by querying the health system's EHR system.As such, any patients without complete information to calculate a LACE+ score or perform CEM were excluded from analysis.This may have skewed the analysis away from patients who received care at an outside institution.However, we routinely obtained a comprehensive medical history and performed a physical examination of each patient during the preoperative neurosurgical office visits.It also remains possible that adverse events were underreported.This limitation was mitigated by an extensive postsurgical follow-up period (mean 429 days).During each office appointment, patients were asked about any additional hospital encounters, including those at outside institutions.
Another source of error may have stemmed from the CEM process.7][28] Other variables, such as the operating surgeon and surgical indications, were not recorded.The binning strategy employed herein was informed by previous studies and our prior experience, but a drawback is that certain sociodemographic variables were analyzed as binary or ternary variables.Because minority groups are at increased risk for surgical complications, further research aimed at isolating the impacts of social determinants of health on patient outcomes is necessary.Nevertheless, because the LACE+ score is a composite score of multiple demographic characteristics, matching may not be necessary to assess its ability to predict outcomes after lumbar fusion.
A final limitation of the present study involves the generalizability of the present findings.Here, we assessed a sample of patients from a single academic medical institution.Furthermore, we did not examine how LACE+ scores changed over time across our institution.Additionally, outcomes were recorded in only the short-term and midterm postoperative windows.Longitudinal, multiinstitutional studies may be warranted to corroborate the results herein.Nonetheless, because lumbar fusion is one of the most common neurosurgical procedures, the present findings remain immediately applicable to numerous neurosurgical patients.

Conclusions
The present study is, to our knowledge, the first to evaluate the efficacy of the LACE+ score in a large, homogeneous population of patients who underwent lumbar fusion.The results suggest that this metric has utility for predicting readmission and other adverse postsurgical events in these patients.Because LACE+ score calculations can be easily incorporated into existing EHR systems, this tool offers an enticing opportunity to seamlessly assess patient risk.

FIG. 1 .
FIG. 1. Flowchart showing the selection of patients who underwent single-level, posterior-only lumbar fusion for analysis and matching.

FIG. 2 .
FIG. 2. Prematch forest plots demonstrating the ORs and 95% CIs of outcomes after single-level, posterior-only lumbar fusion for the sample population (n = 1598) (A), patients with LACE+ scores in quartile 1 (n = 403) compared with patients with scores in quartile 4 (n = 387) (B), and patients with LACE+ scores in the bottom decile (n = 160) compared with patients with scores in the top 4 deciles (n = 672) (C).Red data points indicate significance (p < 0.05).Figure is available in color online only.

FIG. 3 .
FIG. 3. Post-CEM forest plot demonstrating the ORs and 95% CIs of outcomes after single-level, posterior-only lumbar fusion for patients with LACE+ scores in the bottom decile (n = 111) compared with patients with scores in the top 4 deciles (n = 111).

TABLE 1 . LACE+ score sheet
Modified with permission from van Walraven C, Wong J, Forster AJ.LACE+ index: extension of a validated index to predict early death or urgent readmission after hospital discharge using administrative data.Open Med 2012;6(3):e90-e100.CC BY-SA 2.5 (http://creativecommons.org/licenses/by-sa/2.5/ca/).*Scores range from −1 to 90, with higher LACE+ scores predicting greater risk of readmission within 30D after discharge.†Includes age, CCI score, and previous urgent admission.

TABLE 2 . Prematch demographic characteristics of all included patients (n = 1598)
Values are shown as number (%) or mean (range).

TABLE 3 . Post-CEM demographic data of patients with LACE+ scores in the bottom decile (n = 111) and those with scores in the top 4 deciles (n = 111) who underwent single-level, posterior-only lumbar fusion
Values are shown as number (%) or mean (range) unless indicated otherwise.Boldface type indicates statistical significance (p < 0.05).