Perioperative Visual Loss in Spine Fusion Surgery: Ischemic Optic Neuropathy in the United States from 1998 to 2012 in the Nationwide Inpatient Sample

PERIOPERATIVE visual loss (POVL) in spinal fusion surgery is a rare but devastating complication, most commonly due to ischemic optic neuropathy (ION), although other known causes include retinal arterial occlusion (RAO) and cortical blindness. Patients who undergo spinal fusion or cardiac surgery are at the greatest risk for the development of ION.¹  As the utilization of spinal fusion in the United States is the highest in the world,²  it is imperative to uncover the risk factors for and determine how to prevent rare but serious complications such as ION. In a case–control study of 80 subjects with ION after spinal fusion in the American Society of Anesthesiologists (ASA) Postoperative Visual Loss Registry versus unaffected control patients from 17 North American medical centers, we showed that the risk factors for ION included male sex, obesity, the use of a Wilson frame for surgical positioning, anesthesia duration, higher estimated blood loss, and percent colloid of nonblood replacement during the procedure.³ 

The incidence of POVL in spine surgery has been approximated at 1 to 10/10,000.^4,5  Estimates differ widely depending upon size of the study and the data source, e.g., administrative data, single versus multiple institutions, inclusion of all spine surgery, fusions alone, anatomical locations of the surgery, and year of the study, among other factors.³  Although few anesthesiologists and surgeons were familiar with POVL in the 1980s and 1990s, the increasing publication of case reports in the literature,⁶  publicity associated with the ASA POVL Registry,^7,8  and other studies of POVL in the 2000s^9–12  appear to have increased awareness of this complication. Nonetheless, in a survey by the Anesthesia Patient Safety Foundation, 87% of anesthesiologists opined that most surgeons do not recognize the risk of perioperative ION, and 52% believed the same for anesthesia professionals.¹³  There have been other efforts to heighten awareness of POVL in spine surgery among anesthesiologists and surgeons. These included two ASA Advisories on Postoperative Visual Loss, first in 2006 and the other in 2012,^14,15  and a Consensus Statement on Postoperative Visual Loss from the Anesthesia Patient Safety Foundation.¹³ 

Based upon clinical observations, we hypothesized that the incidence of ION in spinal surgery has been decreasing. We also theorized that examination of spinal fusion ION cases in a large, randomly collected database close to the current time would provide new data on pre- and perioperative risk factors. Accordingly, there were two main goals of the current study: (1) to evaluate the incidence trends of ION associated with spinal fusion, and specifically, to determine if the incidence has decreased and (2) to use a large nationwide database to identify risk factors for development of ION in spine fusion surgery.

As there were no patient identifiers, the University of Chicago and University of Illinois Institutional Review Boards deemed the study “exempt.” We studied discharge data in the Nationwide Inpatient Sample (NIS) from 1998 to 2012. The database is maintained by the Healthcare Cost and Utilization Project under the Agency for Healthcare Research and Quality (AHRQ). The hospital discharge record includes patient demographics, diagnoses (principal and less than 14 secondary), procedures (principal and less than 14 secondary), charges (in dollars), length of stay (in days), discharge status, outcomes, and medical diagnoses.^1,16  Diagnoses and procedures are coded using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM).

The NIS is roughly a 20% stratified sample of nonfederal U.S. inpatient hospital discharges and is derived from routine hospital discharge data. In 2012, the sampling was redesigned to improve the accuracy of national estimates; sampling is from all participating hospitals rather than a sample of hospitals. AHRQ provided updated discharge weights on its Internet site for the 1998 to 2012 NIS to ensure accurate weighting of the sample and enable analysis across multiple years.¹⁷  We, thus, used these “trend weights” in the “survey” function of Stata (Stata Corp., USA) for all patient-level analysis.

Quality control and reliability of the NIS have been examined each year since 2000. National estimates of essential healthcare parameters in the NIS were precise and accurate compared with the American Hospital Association Annual Survey, National Hospital Discharge Survey, and Medicare Inpatient data.¹⁸ 

As previously reported,³  ICD-9-CM codes for primary procedure of posterior thoracic, lumbar, or sacral spine fusion in the NIS from 1998 to 2012 were studied, excluding anterior approach to the spine (table 1). Since most ION cases after spinal fusion involve the thoracic and lower vertebrae,⁷  cervical spine fusion was excluded. To ensure complete coverage of relevant procedure codes, ICD-9-CM codes were confirmed against Current Procedural Technology procedure spinal fusion codes using Encoder Pro.com (Optum, USA). Patients discharged with a principal or secondary diagnostic ICD-9-CM code of ION (377.41) were considered to have developed ION during the hospitalization. For comparison of incidence trends, we also identified patients with a diagnosis of retinal artery occlusion (RAO, 362.3; 362.30-0.34, table 1).

Patient characteristics analyzed included age (years, continuous variable), sex, length of hospital stay (days), and yearly inflation-adjusted total hospital charges (both as continuous variables), type of admission (elective, urgent, emergent), discharge status (home, nursing facility, died, etc.), and race. Medical diagnoses (ICD-9 codes are in table 2) studied were atherosclerosis, coronary artery disease, carotid artery stenosis, diabetes mellitus, hypertension, obesity, peripheral vascular disease, and smoking. Hospital conditions included anemia and transfusion. Stroke was included as postoperative, acute, embolic, and thrombotic.

Stata v14.0-MP was used to analyze data. We utilized AHRQ trend weights in the survey function in Stata for all estimates and regressions. Missing data points were noted for race, type of admission, and discharge status. Missing data for race are reported as “unknown” in the regression analysis. Type of admission and discharge status were not included in the regression analysis, and no observations were otherwise excluded due to missing data. To calculate incidence of ION and RAO, we divided the study into 3-yr periods (1998 to 2000, 2001 to 2003, 2004 to 2006, 2007 to 2009, and 2010 to 2012). This enabled the numerator (cases per time period) to reach the threshold for reporting (more than 10), while the denominator was the number of spinal fusion procedures per time period. (AHRQ prohibits data reporting of any data cell, where n is less than or equal to 10.) To study if age was a risk factor, 10-yr age groups were studied, i.e., 18 to 30, 31 to 40 years, etc.

Patient characteristics in spinal fusion groups with or without ION for 1998 to 2012 were tabulated using the national estimates. Univariate analysis, followed by multivariate Poisson regressions, assessed risk factors in the discharges from 1998 to 2012.¹⁹  A Poisson model was used because it was deemed appropriate for modeling count data. P < 0.2 was the minimum criterion for inclusion of variables in the subsequent multiple Poisson regression, and P < 0.05 in the multiple Poisson regression model indicated a significant risk factor. Results are reported as incidence rate ratios (IRRs). The variance inflation factor examined for collinearity.²⁰  Pearson goodness-of-fit test assessed fit of the model.

There were an estimated 2,511,073 discharges in the United States from 1998 to 2012 with procedure codes for posterior thoracic, lumbar, and sacral spinal fusion. The national estimates (table 3) for ION were 257 (1.02/10,000), and for RAO, 224 (0.89/10,000). The volume of procedures per 3 yr increased nearly threefold from 262,279 in 1998 to 2000 to 733,454 in 2010 to 2012 (table 3), but ION in spine fusion (fig. 1) decreased significantly (P = 0.002) from 1.63/10,000 in 1998 to 2000 to 0.60/10,000 in 2010 to 2012. There was no significant change in incidence of RAO during the study period (fig. 1).

Table 4 shows the characteristics in the national estimates of those who sustained ION and those who did not. ION patients were older, and males predominated among ION patients. At least 20% of the race information was not provided. Mean total inflation-adjusted charges for the hospitalization and length of hospital stay were higher for ION than for nonaffected. A higher proportion of ION patients had a blood transfusion (ION, 29.6%; non-ION, 14.0%); the percent of patients with obesity was greater in those with ION (17.1%) than in nonaffected ones (9.8%). Mean number of vascular factors was similar for the two groups; stroke was infrequent, but with higher percent in ION cases.

Poisson multivariate analysis models considered 3-yr interval incidence, age (10-yr intervals), sex, transfusion, obesity, and stroke (those factors with P < 0.2 in univariate analysis, table 5, were entered into the model). Atherosclerosis, carotid artery stenosis, peripheral vascular disease, and race other than white, Hispanic, or Asian/Pacific Islander had insufficient sample size, and therefore, risk with these parameters could not be calculated accurately. In the final model (table 6), the IRR for ION decreased from 1998 to 2012 (IRR, 0.72 per 3-yr period; 95% CI, 0.58 to 0.88; P = 0.002). Factors significantly associated with ION were increasing age (IRR, 1.24 per 10 yr of age; 95% CI, 1.05 to 1.45; P = 0.009), blood transfusion (IRR, 2.72; 95% CI, 1.38 to 5.37; P = 0.004), and obesity (IRR, 2.49; 95% CI, 1.09 to 5.66; P = 0.030). Being female was protective (IRR, 0.30; 95% CI, 0.16 to 0.56; P < 0.000). The Poisson regression showed good fit, as Pearson goodness-of-fit test was not significant at the 5% level. Furthermore, lack of collinearity in the model was suggested by variance inflation factor less than 10 for all covariates.

POVL after spine fusion surgery continues to be a devastating complication, with ION the most frequent cause of visual loss in this surgical cohort. Encouragingly, this study illustrated significantly decreasing ION in spinal fusion; incidence in 2010 to 2012 was about one third than in 1998 to 2000. The lack of change in incidence of RAO suggests that the mechanisms of these disorders are different; moreover, the RAO results serve as an inherent control, further confirming the validity of the results.

We cannot determine the mechanism of this trend in ION. A possible cause is changes in surgical and anesthesia practice, perhaps driven by the heightened awareness generated by the literature, advisories, and consensus statements. However, there are other possible trends that cannot be assessed by our study, such as changes in surgical patient selection, intraoperative positioning, and postoperative management.

The mechanism of perioperative ION, particularly posterior ION, remains elusive, but retrospective studies have revealed associated factors. A case–control study from a single institution examined 126,666 general anesthetic records and found 17 patients with perioperative ION; a subpopulation underwent spine fusion. They could not identify differences in intraoperative hemodynamic parameters between ION patients and the matched controls, suggesting either that patient-specific characteristics may be involved or that the number of cases identified yielded insufficient study power.¹²  Subsequently, a case–control study compared perioperative data between 80 spinal fusion cases with ION from the ASA POVL Registry and 315 unaffected control patients undergoing similar procedures from 17 North American institutions. Male sex, obesity, Wilson frame use, longer anesthetic duration, greater estimated blood loss, and a lower percentage of colloid in nonblood fluid administration were significant and independent risk factors for ION.³ 

In this study, using NIS data, we identified four factors, increasing age, male sex, blood transfusion, and obesity, that were associated with the development of perioperative ION in a multivariate model. While the NIS database does not provide intraoperative data, we demonstrated that two of the factors overlapped with the earlier case–control study: male sex and obesity. Transfusion may be a surrogate for blood loss.

The strengths of this study included a large database, the size of which exceeded that examined in our previous case–control study.³  In the 2012 case–control study, the 80 patients with ION had been anonymously submitted by medical practitioners, quality improvement administrative personnel, and patients. Biased sampling technique or incorrectly entered data could not be ruled out, and some data points were missing in the records.³  For the unaffected controls, the study used a much smaller sampling of hospitals compared to the NIS; hospitals from which the affected patients were derived was anonymous; hence, it is not known if the sampling was representative of all patients in the United States who underwent the procedure. But, the NIS has the advantage of providing a randomized sampling of discharges from hospitals.

Why men were at higher risk for ION remains unclear since there are no known anatomic variations between the male and female visual pathways. An influence of female hormones including estrogen could play a role.²¹  Acute venous congestion of the head and neck secondary to patient position is a possible mechanism for ischemic damage to the optic nerve, which corresponds well with one of the theories on ION pathogenesis.²²  Anatomic differences between men and women, which may become evident when they are positioned on the Wilson frame, may play a role in causing greater interstitial fluid accumulation in men that may predispose to a higher incidence of ION. It is possible that obese men are more likely to lose greater amounts of blood and require more fluid during spine surgery.

A new finding was that age was a risk factor for ION in spinal fusion. There was a 24% increase in incidence risk ratio per 10 yr of age. It may be related to older patients requiring more complex surgery or to increased vulnerability of the optic nerve to physiologic insults associated with major spine surgery, as the optic nerve undergoes significant degeneration with age.²³  We found no influence of diabetes mellitus, hypertension, coronary artery disease, number of vascular risk factors, or smoking on the risk of developing ION. Stroke (none were postoperative stroke) was not significant in the multivariate analysis. However, sample sizes of affected patients were small and could have influenced these results. Due to missing data, it is not possible to assess the impact of any possible racial disparities on the results.

Our study demonstrated longer length of stay and higher hospital charges in patients who sustained ION. While these patients typically undergo more in-hospital diagnostic procedures, and/or treatment of ION, we cannot definitively conclude that longer stay and higher charged were caused by ION. Patients who are older and undergo more complex surgery, e.g., greater number of spine levels fused, and hence sustain longer hospital stays, could also be the same patients who are at risk of ION. However, we were unable to study the effect of number of levels fused as these data were only sparsely present in the database.

In addition to understanding the risk factors associated with ION, there has been a coordinated and concerted effort to increase awareness and help guide patient management. In 2006 (and revised in 2012, but the latter cannot be related to the results in this study), the ASA produced practice advisory statements to help educate and guide practice for spine fusion surgery.¹⁵  We cannot determine from the current study what role, if any, these advisory statements have had on the changing incidence of ION. One notable change in spine surgery has been the increasing use of “minimally invasive” surgical techniques.²⁴  These have been shown to result in less blood and fluid requirements compared to the traditional “open” approach.^24,25  Although it would have been relevant to study the impact of this change on ION incidence, there was no specific procedure code for the “minimally invasive” procedure before 2013.²⁶ 

There are limitations to this study. The NIS database relies on the accuracy of diagnosis and coding. Verification of each individual case and the associated medical diagnoses are not possible as the discharge records are deidentified. Currently, there are not enough data in the literature to assess the accuracy of coding in NIS for uncommon complications such as ION and RAO.²⁷  Both over- and undercoding are possible. Also, it is not possible to rule out if changes in coding frequency of ION or of the spinal fusion procedure codes are involved in the incidence data results. The coding discharge information depends upon entry of the data by professional coders; however, the accuracy of the information in turn depends upon the diagnoses recorded by physicians and the procedure description provided by the surgeon. To avoid confounding issues with procedure coding, a wide range of codes for posterior spinal fusion was used and compared to Current Procedural Technology codes, which are more commonly used for procedure coding in the United States for billing purposes. However, given the nationwide scope of the NIS, systematic bias in this coding appears improbable. Some cases of ION may have been preexisting which would lead to a false elevation of the incidence. The severity of injury using the NIS cannot be quantified, and the type of ION (anterior vs. posterior) cannot be differentiated. The NIS only identifies diagnosed cases of ION but does not stratify the extent of loss of vision. NIS data lack longitudinal information (i.e., a single individual cannot be tracked across multiple hospitalizations or for follow-up after surgery). NIS does not contain intraoperative data such as anesthetic technique, length of surgery, surgical positioning, blood loss, or volume of fluids infused.¹⁶ 

Definitive conclusions concerning changes in intraoperative technique and practice cannot be made from this study. This area remains challenging to appropriately study in the clinical setting since rates of ION are low and deliberately manipulating intraoperative variables in a randomized controlled design would be difficult if not unethical. Despite the continual and dramatic increase in spinal fusion utilization, the rates of ION after spine fusion in our study have declined significantly between 1998 and 2012. Further study of large multicenter databases that contain more pertinent intraoperative data may allow assessment of whether perioperative surgical and anesthetic practice has been modified consistent with the recommendations of the national practice advisories.

The authors are grateful to Ms. Elaine Hrinyo, C.P.C., Billing Manager, Department of Anesthesia and Critical Care, University of Chicago Medicine, Chicago, Illinois, for assistance with Current Procedural Technology codes, and the Encoder program.

Supported by National Institutes of Health (Bethesda, Maryland) grants RO1 EY10343 (to Dr. Roth), UL1 RR024999 (to the University of Chicago Institute for Translational Medicine, Chicago, Illinois), K23 EY024345 (to Dr. Moss), core grant P30 EY001792 (to the Department of Ophthalmology, University of Illinois at Chicago, Chicago, Illinois), and an unrestricted grant from Research to Prevent Blindness (New York, New York) to the Department of Ophthalmology and Visual Sciences, University of Illinois at Chicago.

Drs. Roth and Lee have served as expert witnesses in cases of perioperative eye injuries on behalf of patients, physicians, and hospitals. The other authors declare no competing interests.