Secondary Logo

Journal Logo

ORIGINAL ARTICLES

International Classification of Disease Clinical Modification 9 Modeling of a Patient Comorbidity Score Predicts Incidence of Perioperative Complications in a Nationwide Inpatient Sample Assessment of Complications in Spine Surgery

Chitale, Rohan MD*; Campbell, Peter G. MD*; Yadla, Sanjay MD*; Whitmore, Robert G. MD; Maltenfort, Mitchell G. PhD; Ratliff, John K. MD§

Author Information
Journal of Spinal Disorders & Techniques: May 2015 - Volume 28 - Issue 4 - p 126-133
doi: 10.1097/BSD.0b013e318270dad7
  • Free

Abstract

Current efforts to improve the quality of health care revolve around attempting to reduce incidence of hospital-acquired conditions and perioperative adverse events. Strategies include limiting reimbursements for patients who develop hospital-acquired conditions and “pay for performance” plans in which practitioner’s payment is linked to patient outcome. The “pay for performance” guidelines also reward providers for reaching higher performance standards in predefined areas, and thus aim to effect change in physician practices to attain improved health care.1 Variability in patient condition and in relevant patient comorbidities likely plays a role in complication incidence and overall clinical outcome. Financial impact of these quality improvement measures may be significant.1

Prior studies have reported a wide range of incidence of complications in spine surgery, with retrospective reviews consistently underreporting complication incidence.2 Our study group previously completed a prospective assessment of complications in spine surgery which led to development of a basic comorbidity score (RCS). We have previously presented this measure as a predictor for risk of complication occurrence for patients undergoing spine surgery procedures.3

To aid in validation of our comorbidity score, we developed an International Classification of Disease Clinical Modification (ICD-CM) 9 model of the score, choosing families of diagnostic codes to represent the individual comorbidities assessed in the RCS. Development of an ICD-9-CM model of the score will facilitate validation of the score through application to large clinical databases and also will allow for comparison with other clinical measures predicting complication occurrence. Similar approaches to assessing spine surgery complications in population databases has been previously used and validated.4–6

The goal of this study is to validate the previously reported RCS through assessment of a much larger group of patients through ICD-9-CM modeling and to assess the accuracy of the RCS by comparison with a similarly modeled ICD-9-CM standard for assessment of perioperative complication risk, the Deyo modification of the Charlson comorbidity score.

MATERIALS AND METHODS

Study Design

A prospective assessment of perioperative complications in patients undergoing spine surgery was completed. Results of this assessment have been previously reported.7–9 We developed a comorbidity score through assessing the most common comorbidities in our patient population and comparing them with risk of developing a complication.3

We defined complications on the basis of a survey of spine surgeons and spine surgery patients.10,11 In the prospective assessment, perioperative complications were parsed into major and minor groups. Major complications were those that resulted in permanent sequelae or further operative intervention, and include infection, pneumonia, renal failure, Myocardial infarction (MI), pulmonary complications, neurological complications, congestive heart failure, pulmonary embolus, hematoma, and wound complications. Minor complications were those that were transient and did not require further intervention, including urinary tract infection, cardiac dysrhythmia, seroma, delirium, and deep venous thrombosis. Patients undergoing cervical spine surgery were examined separately from those undergoing thoracolumbar spine surgery. A comorbidity score was calculated to predict complication rate in these patients.3

Nationwide Inpatient Sample (NIS)

The Healthcare Cost and Utilization Project NIS is a database containing inpatient data from over 1000 hospitals within the United States. NIS and other databases have been widely used to assess spine surgery procedures and patient factors, including complication incidence.4,6,12,13

A database cohort of patients undergoing elective spine surgery from 2002 to 2009 was collected from the NIS database. All patients 18 years or older with ICD-9-CM codes indicating degenerative cervical or thoracolumbar spine disease and associated surgery were included in the study. Both preoperative comorbidities and perioperative complications were collected in this patient sample. Table 1 reviews a summary of the definitions of comorbidities used in this study. In this retrospective assessment, we used ICD-9 codes to compile general complication statistics.7 We did not separate major and minor complications because of previously described limitations of the ICD-9-CM coding information used in the NIS database.7

TABLE 1
TABLE 1:
Codes Used to Determine Initial Patient Population

Analysis was limited to elective procedures completed for degenerative pathologies, limiting the potential for bias from inclusion of emergent cases and nondegenerative pathologies (Table 1). We omitted all patients treated for other pathologies, realizing the potential bias in complication incidence implicit in differing patient diagnoses.8,9

The RCS was originally based upon assessment of the most common comorbidities present in a prospective cohort of spine surgery patients. We developed an ICD-9-CM model for the previously developed comorbidity score, with ICD-9-CM codes mapping the individual aspects of the score (Table 2). Healthcare Cost and Utilization Project maps comorbidity codes into subheadings for reporting in the NIS; where appropriate these subheadings were utilized and are included in Table 2. To calculate the RCS, 1 single point is given for each preoperative comorbidity present. In contradistinction to the Charlson index, the RCS uniformly weighs each comorbidity category to simplify scoring.

TABLE 2
TABLE 2:
Comorbidities Used to Compile the RCS, With Relevant ICD-9-CM Codes

The Charlson score was compiled as described by Deyo et al.14 Codes chosen for mapping of the Deyo modification of the Charlson comorbidity score are reviewed in Table 3. The Charlson index separates mild and severe liver diseases into 2 categories, assigning mild disease a value of 1 and severe disease a value of 3. Because the NIS database did not differentiate between these 2 categories, the diagnosis of “liver disease” contributed a value of 2 to our modified Charlson index. We assessed the predictive capacity of the comorbidity score in comparison with the Charlson comorbidity score.

TABLE 3
TABLE 3:
Codes Used in Calculating Deyo Modification of Charlson Comorbidity Score

Assessment of Comorbidities/Complications

After tallying the comorbidities for each patient group, we sought to compare 2 comorbidity indices to evaluate their ability to predict complications in patients undergoing spine surgery. Codes used to denote complications are reviewed in Table 4.

TABLE 4
TABLE 4:
ICD-9-CM Codes Used for Complication Assessment

Statistics

Data were analyzed using the R statistical software (Vienna, Austria). Probability values of <0.05 were considered statistically significant. Analysis was conducted on the patient cohort with respect to the Charlson and the RCS indices. ORs were calculated to determine the strength of association between independent factors and complication rate. These factors included Charlson index, age, and individual comorbidities.

For the RCS, multivariate models were developed by first creating a full model with all feasible explanatory variables and then systematically removing the least significant variables until a minimal model was reached. Logistic regression was used to predict the probability of occurrence of complications based on combinations of comorbidities. In this way, the logistic regression allowed for each comorbidity to be given a score which it contributed to the RCS. The nomogram function in the RMS package for R was used for this computation. The comorbidity categories used in the calculation of the RCS included neurological deficit, cardiac condition, drug or alcohol use, chronic pulmonary condition, diabetes mellitus, psychiatric illness, sex, age, and presence of hypertension. MI was not included in this list because, based on the NIS compilation of data, we could not differentiate between preexisting history of a comorbidity versus in-hospital conditions suffered in the perioperative period.15 However, “cardiac conditions,” which were included as a comorbidity category, incorporates conditions related to or leading to an MI. Therefore, the MI comorbidity was implicit in the cardiac comorbidity group. In contrast, “neurological deficit” was included even though this distinction between preoperative comorbidity and perioperative condition could not be made because “neurological deficit” as a comorbidity was not implicit in any other group.

The RCS was then compared with the Charlson index to evaluate ability to predict complications in our patient cohort undergoing spine surgery. Cervical spine patients were considered separately from thoracolumbar patients. Comparison was made by developing a receiver-operating curve (ROC) for each index. The ROC is a graphical plot of the true-positive rate (sensitivity) versus the false-positive rate (1 specificity) for a series of detection thresholds to evaluate whether the index correctly predicts complication rate. It is developed by changing the threshold for which the index predicts a complication and evaluating how the sensitivity and specificity of the index changes. As the threshold value increases, the index should yield fewer true positives(less sensitive) and fewer false positives (greater specificity). Therefore, the best possible ROC curve would yield a point at (0, 1) in the ROC space, where the true-positive rate is 1.0 and the false-positive rate is 0. A diagonal line from (0, 0) to (1, 1) in the ROC space would yield a “line of no discrimination” where any point on the line represents a random guess. The area under the ROC curve represents its accuracy, or the probability that the index will correctly identify the occurrence of a complication. An ROC curve with an area=1 reflects the ideal curve, whereas a diagonal ROC curve of the line of no discrimination, with an area of 0.5, represents a pure guess (flip of the coin). Thus, an ROC line above the line of no discrimination represents better performance of the index, whereas a line at or below the line of no discrimination represents worse performance. This ROC curve was calculated for both the RCS and the Charlson index in both the cervical and thoracolumbar cohorts.

RESULTS

A total of 352,535 patients undergoing 369,454 spine procedures for degenerative disease were gathered from the NIS database. There were 184,127 cervical patients and 185,327 thoracolumbar patients. Female patients comprised 87,292 (52.18%) of the cervical group and 101,769 (54.94%) of the thoracolumbar subset. Overall, the cervical procedures resulted in 8286 complications (4.50%) whereas the thoracolumbar procedures produced 25,118 complications (13.55%).The median age bracket of both cervical and thoracolumbar patients was 45–50 years (Fig. 1). Table 5 reviews patient demographics and the incidences of relevant comorbidities.

FIGURE 1
FIGURE 1:
Shows the age distribution for the patient populations reported. The x-axis denotes age categories in 5-year increments, the y-axis represents number of patients in each age group recorded in the assessment. NIS indicates Nationwide Inpatient Sample; TL, thoracolumbar.
TABLE 5
TABLE 5:
Patient Demographics, NIS ICD-9-CM Sample

Data for 25 different comorbidity categories were collected. Hypertension was the most common comorbidity among cervical patients, totaling 64,863 (35.23%). Hypercholesterolemia represented the most common comorbidity for thoracolumbar patients, with 143,282 (77.31%) patients affected.

We calculated the RCS for each patient in the cohort and compared this value with complication incidence. Increasing RCS correlated linearly with increasing complication incidence (Fig. 2; OR 1.11; 95% CI, 1.10–1.13; P<0.0001).

FIGURE 2
FIGURE 2:
Shows relationship between the derived comorbidity score (RCS; x-axis) and complication incidence (y-axis, %). There was a clear relationship between increasing number of comorbidities and increasing incidence of perioperative complications.

Logistic regression was performed to elucidate each comorbidity’s contribution to complication rate. In both the cervical and thoracolumbar patient cohorts, neurological deficit, cardiac conditions, and drug or alcohol use had the highest strength of association with complication rate. Odds ratios for specific comorbidities comprising the RCS are listed in Table 6. Although not all individual factors used in deriving the RCS were found to correlate with increased complication incidence, all comorbidities denoted in Table 2 were included in RCS calculation.

TABLE 6
TABLE 6:
Impact of Individual Factors Contributing to the RCS

Different factors were found individually significant in different spinal regions. Although neurological disorders, alcohol and drug abuse, and cardiac disorders other than hypertension consistently correlated with high complication incidence, systemic malignancy was not correlated with increase in cervical complication incidence. It is of note that our restriction to solely degenerative patients eliminates surgical treatment of neoplastic spinal pathology.

The ICD-9-CM model of the Charlson comorbidity score also revealed correlation between increase in score and increase in complication incidence. The estimated OR for each point increase in the Charlson index was 1.25 (95% CI, 1.23–1.27) in the cervical patient group and 1.11 (95% CI, 1.10–1.12) in the thoracolumbar group.

The ROC curve was plotted for the RCS and the Charlson indices in the cervical patient cohort as well as in the thoracolumbar patient cohort (Figs. 3, 4). In both groups, the RCS and Charlson indices produced curves that are situated to the left of the line of no discrimination, demonstrating their predictive value. The area under the curve, representing the accuracy of the index, was 0.66 for the RCS and 0.65 for the Charlson index in the cervical patient subgroup. The area under the curve was 0.60 for the RCS and 0.59 for the Charlson index in the thoracolumbar patient subgroup. This indicates that the RCS performs at least as well as the Charlson index in predicting complication occurrence in both cervical and thoracolumbar spine patients.

FIGURE 3
FIGURE 3:
Shows the receiver-operating curve (ROC) assessment of the RCS and Charlson comorbidity scores for cervical patients. Increase in each was found to predict complication incidence in elective cervical cases.
FIGURE 4
FIGURE 4:
Shows a similar receiver-operating curve (ROC) assessment of the RCS and Charlson comorbidity scores for thoracolumbar patients. Again, increase in each scale predicted complication incidence.

DISCUSSION

The rate of complications in spine surgery may be impacted by a variety of factors including pathology, surgical approach, surgical technique, and patient comorbid conditions. Increasing number of comorbidities may result in increased risk of perioperative adverse events. In this study, RCS correlated with complication occurrence, with increasing score implying increasing risk of perioperative complications in an elective group of spine surgery admissions for degenerative pathologies. The RCS performed at least as well as the Charlson index in predicting risk of complication in spine patients when tested restrospectively with the NIS database. The RCS may be a useful tool to help stratify risk of complications due to comorbid conditions for an individual undergoing elective spine surgery.

Prospective Study

Our study group previously completed a prospective analysis of complications occurring in a cohort of spine surgery patients. Through a prospective analysis, we sought to obtain a more accurate measure of complication incidence and develop a relative risk model to define the relationship of complication incidence to patient factors. These findings have been previously reported.7–9,11 Our previous assessments have revealed that patient comorbidities significantly increase the risk of perioperative complications, with that risk amplified as the number of comorbidities in an individual patient increases.3

A comorbidity score was developed by assessing the most common comorbidities present in a prospectively assayed group of spine surgery patients and then quantifying the impact of increasing number of comorbidities on complication incidence. The 11 most common comorbidities found in our prospectively accumulated patients were chosen for incorporation in the RCS. No weighing the individual comorbidities was attempted because of the small sample size of the initial patient cohort. In our prospective patient population, we found correlation between increasing number of comorbidities and increasing risk of perioperative complications.3

Charlson Score

The Charlson comorbidity index was designed to predict a 10-year mortality rate for any patient based on his comorbid conditions.16 It is calculated by assessing a group of patient comorbidities, assigning each condition a value (1, 2, 3, or 6), and adding them together to give a total score. Codes chosen for mapping of the Deyo modification of the Charlson comorbidity score are reviewed in Table 3. The Charlson score, although designed to predict mortality in cancer patients, has been used to predict mortality after spine trauma and morbidity after surgical intervention.17,18

Need for Relative Risk Model

Success in delivering health care requires optimizing outcomes and minimizing adverse events. In spine surgery, achieving this goal entails proper selection of surgical candidates. In a prior study we prospectively assessed the impact of comorbidities on the rate of complication in patients undergoing spine surgery. We found that relative risk of complications increased with the number of comorbidities in an individual patient.3 Other retrospective and NIS-based studies of selected spine patients have also shown greater risk of perioperative complications with increasing number of comorbidities and advancing patient age.19–21

A previous assessment of our prospectively accrued patient cohort explored the capacity of the aspirin physical status score and a prospectively measured Deyo modification of the Charlson score to predict complication incidence. Increasing aspirin score was sensitive for predicting occurrence of major complications, but did not correlate with overall complication incidence. Increase in the Charlson score paralleled overall and minor complication incidence, but did not correlate with occurrence of major complications (Whitmore/Ratliff, personal communication). On the basis of this analysis of our prospective patient population, it is evident that a better means of predicting complications in spine surgery patients would be desirable.

Results of Present Study

We used the NIS database to retrospectively identify 352,535 patients who underwent elective surgery for degenerative cervical or thoracolumbar spine disease. The comorbidites for each patient were noted along with any perioperative complications. We determined the modified Charlson comorbidity index as well as the RCS for each patient. Analysis of the database revealed that both the indices were able to predict risk of complications with some success. Using an ROC to measure accuracy of the different relative risk models, the RCS was at least as successful as the Charlson index in predicting complication occurrence in our patient population (Figs. 3, 4).

Complication incidence was considerably lower than previous prospective reports generated by our group. This is likely secondary to consideration of only events occurring during the studied patient population’s original hospital admission, hence missing readmissions for adverse events manifesting after discharge, and the established lower accuracy of ICD-9-CM database capture of perioperative complications.7 Our analysis proves that the simple comorbidity score performs at least as well as the more cumbersome Charlson index in determining relative risk of perioperative complications in this population of degenerative spine surgery patients. The predictive ability of the RCS is validated in this retrospective database study.

Opportunities for Bias

There are potential areas of bias in this study. We gathered comorbidity and complications data based on ICD-9-CM codes in the NIS registry. Our ability to determine complication incidence and variety is dependent on the comprehensiveness and accuracy of the database itself.7

Our definition of comorbidity was broad, encompassing 25 categories of ICD-9-CM coding. The wide-ranging pathologies that we included from the NIS database, including some not used by other authors, may have led to incorporation of comorbidities into the RCS with limited clinical impact.22–24 Restricting the comorbidities included in the RCS to the most common comorbidities seen in the prospective analysis reduces this potential bias. Logistic regression analysis also enabled us to stratify the impact of each comorbidity.

In determining the Charlson comorbidity index, we were again limited by the comorbidity classifications of the NIS database. We determined the Charlson score as described by Deyo, with modifications due to the inadequacy of the NIS database to account for severity of comorbid conditions such as liver disease.14 While paralleling the approach taken by other authors, these liberties with the NIS dataset introduce another opportunity for bias in our results.23

In some diagnostic categories, the database also did not allow differentiation of preoperative comorbid conditions from perioperative complications. For example, the ICD-9 code for myocardial infarction did not carry with it a modifier to discriminate between a preoperative history and perioperative complication of myocardial infarction. This weakness of the NIS dataset may reduce the accuracy of the analysis.15

Because disease pathologies impact our patient population differently, we restricted the patient sample to include those solely treated for degenerative spine disease. Our goal was to limit the bias in complication incidence that occurs from different diagnoses.8,9 However, by restricting our assessment to only degenerative patients, we limit the relevance of these results to assessment of other patients undergoing spine surgery.

CONCLUSIONS

This study validates the RCS as an effective means of predicting complication occurrence in elective cervical and thoracolumbar spine surgery patients. The RCS proved at least as effective as the widely used and more cumbersome Charlson comorbidity score. Neither measures were completely accurate, however; further work is needed in developing a more robust relative risk modeling of complications in spine surgery procedures. Future study will incorporate a broader assessment of spine surgery patients, incorporating nondegenerative pathologies, and also explore weighing of the RCS on the basis of contribution of individual comorbidities to complication occurrence.

REFERENCES

1. Teufack SG, Campbell P, Jabbour P, et al.. Potential financial impact of restriction in “never event” and periprocedural hospital-acquired condition reimbursement at a tertiary neurosurgical center: a single-institution prospective study. J Neurosurg. 2010;112:249–256.
2. Nasser R, Yadla S, Maltenfort MG, et al.. Complications in spine surgery. J Neurosurg Spine. 2010;13:144–157.
3. Campbell P, Yadla S, Nasser R, et al.. Patient comorbidity score predicts incidence of perioperative complications: Assessing the impact of patient comorbidities on complications in spine surgery. J Neurosurg Spine. 2011;16:37–43.
4. Wang M, Chan L, Maiman D, et al.. Complications and mortality associated with cervical spine surgery for degenerative disease in the United States. Spine. 2007;32:342–347.
5. Martin B, Mirza S, Comstock B, et al.. Are lumbar spine reoperation rates falling with greater use of fusion surgery and new surgical technology? Spine. 2007;32:2119–2126.
6. Deyo R, Mirza S, Martin B, et al.. Trends, major medical complications, and charges associated with surgery for lumbar spinal stenosis in older adults. JAMA. 2010;303:1259–1265.
7. Campbell PG, Malone J, Yadla S, et al.. Comparison of ICD-9-based, retrospective, and prospective assessments of perioperative complications: assessment of accuracy in reporting. J Neurosurg Spine. 2011;14:16–22.
8. Yadla S, Malone J, Campbell PG, et al.. Early complications in spine surgery and relation to preoperative diagnosis: a single-center prospective study. J Neurosurg Spine. 2010;13:360–366.
9. Yadla S, Malone J, Campbell PG, et al.. Preoperative diagnosis and early complications in thoracolumbar spine surgery: a single center prospective study. J Spinal Disord Tech. 2010;24:E16–E20.
10. Lebude B, Yadla S, Albert T, et al.. Defining “complications” in spine surgery: neurosurgery and orthopedic spine surgeons’ survey. J Spinal Disord Tech. 2010;23:493–500.
11. Ratliff JK, Lebude B, Albert T, et al.. Complications in spinal surgery: comparative survey of spine surgeons and patients who underwent spinal surgery. J Neurosurg Spine. 2009;10:578–584.
12. Gray D, Deyo R, Kreuter W, et al.. Population-based trends in volumes and rates of ambulatory lumbar spine surgery. Spine. 2006;31:1957–1963.
13. Deyo R, Gray D, Kreuter W, et al.. United States trends in lumbar fusion surgery for degenerative conditions. Spine. 2005;30:1441–1445.
14. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992;45:613–619.
15. Roos LL, Stranc L, James RC, et al.. Complications, comorbidities, and mortality: improving classification and prediction. Health Serv Res. 1997;32:229–238discussion 239–242.
16. Charlson M, Szatrowski TP, Peterson J, et al.. Validation of a combined comorbidity index. J Clin Epidemiol. 1994;47:1245–1251.
17. Arrigo R, Kalanithi P, Cheng I, et al.. Charlson score is a robust predictor of 30-day complications following spinal metastasis surgery. Spine. 2011;36:E1274–E1280.
18. Harris M, Reichmann W, Bono C, et al.. Mortality in elderly patients after cervical spine fractures. J Bone Joint Surg Am. 2010;92:567–574.
19. Kalanithi PS, Patil CG, Boakye M. National complication rates and disposition after posterior lumbar fusion for acquired spondylolisthesis. Spine (Phila Pa 1976). 2009;34:1963–1969.
20. Li G, Patil CG, Lad SP, et al.. Effects of age and comorbidities on complication rates and adverse outcomes after lumbar laminectomy in elderly patients. Spine (Phila Pa 1976). 2008;33:1250–1255.
21. Cloyd JM, Acosta FL Jr, Cloyd C, et al.. Effects of age on perioperative complications of extensive multilevel thoracolumbar spinal fusion surgery. J Neurosurg Spine. 2010;12:402–408.
22. Veeravagu A, Patil CG, Lad SP, et al.. Risk factors for postoperative spinal wound infections after spinal decompression and fusion surgeries. Spine (Phila Pa 1976). 2009;34:1869–1872.
23. Harris MB, Reichmann WM, Bono CM, et al.. Mortality in elderly patients after cervical spine fractures. J Bone Joint Surg Am. 2010;92:567–574.
24. McHenry TP, Mirza SK, Wang J, et al.. Risk factors for respiratory failure following operative stabilization of thoracic and lumbar spine fractures. J Bone Joint Surg Am. 2006;88:997–1005.
Keywords:

complications; comorbidities; spine surgery; outcomes

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.