Corneal Abrasion in Hysterectomy and Prostatectomy: Role of Laparoscopic and Robotic Assistance

PROSTATE cancer, the most common non-skin male cancer in the United States, is also the second most frequent cancer death. Radical prostatectomy (RP) is conventional for clinically localized cancer. Robotic-assisted laparoscopic radical prostatectomy (R/PROST) has advantages of shortened hospital stay and decreased blood loss, morbidity, and postoperative analgesia compared with open prostatectomy (OP).¹ 

Hysterectomy has been performed for decades, for both malignant and benign disease, and in one in three women by age 60 yr.²  With faster recovery, less pain, and fewer overall complications, robotic-assisted laparoscopic hysterectomy (R/HYST) has increased dramatically, although without clear benefit for benign disease.³  In malignancy, initial outcomes of robotic surgery are encouraging.^4,5 

Retro- or prospective studies of anesthetic complications are lacking in R/HYST, but reported challenges of R/PROST are altered respiratory mechanics, laryngeal edema, and brachial plexus injuries.¹  In a small study of 1,500 R/PROST patients, corneal abrasion (CA) was the most reported anesthesia-related complication (3%).⁶  Because R/HYST and R/PROST share steep head down positioning, pneumoperitoneum, and a surgical learning curve, among other similarities, we hypothesized a similar incidence of eye-related complications.^7,8 

We aimed to determine the incidence of CA after prostatectomies and hysterectomies, to test the hypothesis that laparoscopic or robot assistance increases risk, and to examine risk factors. Rapid introduction of the robot for RP rendered it difficult to study the influence of laparoscopy versus robotic assistance on CA. However, hysterectomy, a similar procedure, is easily identifiable as open, laparoscopic, or robotic assisted. We used the Nationwide Inpatient Sample (NIS), a large administrative discharge database that allows study of low-frequency events.⁹  CA is painful and disturbing and has been targeted for anesthesia performance improvement as potentially preventable.¹⁰  With robotic-assisted surgery increasing, it is important to understand the increased risks of these unexpected and painful perioperative complications.

Data were extracted for 2000–2011 from the NIS, the largest United States all-payer hospital inpatient database. With over 1,000 randomly selected hospitals in 44 states, it is an approximately 20% stratified sample, including public hospitals and academic medical centers. As part of the Healthcare Cost and Utilization Project, NIS is maintained by the Agency for Health Care Research and Quality. Hospitals and discharges are weighted based on the number of hospitals and discharges in the database.^*

A typical hospital discharge includes patient demographics (age, sex, race), diagnoses (principal and <14 secondary), procedures (principal and <14 secondary), charges, length of stay, discharge status, outcomes, and number of chronic conditions.^† Diagnosis and procedure data are coded using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). Quality control and reliability of the NIS have been examined each year since 2000.^‡ National estimates of essential health care parameters in the NIS were precise and accurate compared with the American Hospital Association Annual Survey, National Hospital Discharge Survey, and Medicare Inpatient data.^§ Our Institutional Review Board deemed the study “exempt.”

Discharges with RP, open hysterectomy (O/HYST), or laparoscopic hysterectomy (L/HYST) were identified using ICD-9-CM procedure codes (table 1) and confirmed with Encoder Pro.com (Optum, USA). While the L/HYST codes were specific, to identify laparoscopic RP required modifier code 51.42. However, most of the RPs in NIS from 2009 to 2011 were either OP or robotic assisted, with few identifiable as laparoscopic, without robotic assistance. Because the number of laparoscopic, non–robotic-assisted RPs in NIS was only 376, we eliminated these for insufficient sample size (fig. 1). As recognized by the National Center of Health Statistics and the Centers for Medicare and Medicaid Services, beginning October 1, 2008, the robot-assisted modifier codes (ICD-9-CM 17.42 and 17.44) were introduced to specifically identify R/PROST as well as R/HYST. Thus, discharges with procedure codes for both RP and the robot-assisted modifiers were classified as R/PROST, whereas those that only had the procedure code for RP were classified as non-R/PROST, that is, OP (table 1 and fig. 1). L/HYST discharges that contained 17.42 or 17.44 codes were classified as R/HYST. Because the robotic modifier code was introduced in late 2008, we limited robotic modifier codes to 2009–2011. At the time of our analysis, 2011 was the most current database. CA was identified as ICD-9-CM 371.20.

We extracted all discharges containing R/PROST and non-R/PROST, R/HYST, L/HYST, and O/HYST in 2009–2011 and those that also had discharge codes corresponding to CA. We calculated CA incidence and the number needed to harm (NNH); also, we compared age, race, and number of chronic medical conditions between the groups, as we previously reported.⁹  Odds ratios (ORs) were calculated by univariate and multivariate analysis. Moreover, within univariate and multivariable analysis, we assessed R/HYST and L/HYST within a categorical variable “Modified Hysterectomy” having two levels for each procedure variant. For hysterectomy, cases were segregated as malignant versus nonmalignant using ICD-9-CM codes: 180 (carcinoma of the cervix), 182 (carcinoma of the uterus), and 183 (carcinoma of the fallopian tubes and broad ligament).

Data were analyzed with STATA v13.0 (Stata Corporation, USA). All R/PROST and OP discharges, and separately, those discharges with diagnostic codes for CA, were compared between groups for mean age and number of chronic conditions using an independent sample, 2-tailed t test and proportions of distribution according to race (white, African American) using chi-square, and P less than 0.05 significant. For R/HYST, L/HYST, and O/HYST, three-way comparisons used ANOVA with Bonferroni correction for three comparisons to compare mean age and mean number of chronic conditions; P less than 0.05 was significant. Chi-square analyses assessed race distributions and the presence of malignancy, and were corrected for multiple pairwise comparisons with P less than 0.05/3 = 0.017.

To examine factors increasing CA risk in RP and hysterectomy, we first used univariate analysis. As prostatectomy and hysterectomy are distinct procedures on opposite genders, we maintained separate analyses. Age, number of chronic conditions, race, surgery year, malignancy diagnosis, and robotic or laparoscopic procedure were examined. For univariate analysis, P less than 0.2 was the cutoff for entering parameters into the subsequent multiple regression analysis.

Collinearity tests were performed on remaining candidate parameters. Continuous, increasing variables including age and number of chronic conditions were compared through Pearson correlations, and discrete variables including malignant, robotic, laparoscopic, and race were compared using chi-square. Unpaired t tests compared continuous and discrete variables. Correlations above 0.5 with P less than 0.05 were deemed significant, as were P values less than 0.05 for unpaired t tests and chi-square tests.

Multiple logistic regression was performed using procedure modifiers, and the additional covariates age, number of chronic conditions, malignancy, race, and year in RP, and separately in hysterectomy. These models thus considered the impact of surgical technique on CA in the context of all the background covariates, effectively controlling for differences between group compositions. Forward and backward elimination yielded the same results. An area under the receiver operating characteristic (ROC) curve greater than 0.5 and Hosmer and Lemeshow P greater than 0.05 were the thresholds for confirming the goodness of fit and validity of these models, respectively.

NIS contained 166,942 RP discharges from 2000 to 2011, with 295 CAs (0.18%, 1.8 per thousand; table 2). The total number of CAs during this same time period increased; in 2000, there were less than 10 discharges with CA (no more than 0.08%, 0.8 per thousand), whereas in 2011 there were 46 (0.28%, 2.8 per thousand), an increase of approximately three-fold. (NIS does not permit reporting of research results with individual values <10, hence, these are expressed as “<10.”) We identified 216, 890 L/HYST discharges from 2000 to 2011, among them were 275 with CA, 0.13%, or 1.3 per thousand. There were less than 10 CAs in 2000 (≤0.09%, or 0.9 per thousand), and 55 (0.21%, 2.1 per thousand), an approximate two-fold increase, in 2011 (table 2). The rate of CA in O/HYST remained stable during 2000–2011, an overall rate of 0.03%, or 0.3 per thousand (table 2).

In 2009–2011, of 28,521 discharges with R/PROST, there were 99 CAs (3.5 per thousand); while 17,554 discharges with OP had 34 CAs (1.9 per thousand, table 3). Among all discharges with R/PROST or OP, the latter had a greater mean number of chronic conditions (P < 0.0001 by unpaired t test) and a higher proportion of African Americans (P < 0.0001 by chi-square). Patients undergoing R/PROST and OP who developed CA were not significantly different in age, number of chronic conditions, or race (table 3).

In hysterectomy, among 191,199 O/HYST, there were 70 CAs (0.4 per thousand) while in L/HYST, there were 70 in 51,323 (1.4 per thousand), and when robotic assistance (R/HYST) was used, CA was in 63 of 21,123 (3.0 per thousand, table 3). R/HYST patients were significantly older and had greater average number of chronic conditions versus O/HYST and L/HYST and contained a lower proportion of African Americans compared with O/HYST (table 3). Moreover, L/HYST discharges were significantly younger, had lower average number of chronic conditions, had a lower proportion of African Americans, and had lower proportion of malignancy compared with O/HYST. R/HYST had significantly higher proportions of malignancy diagnoses compared with the other two groups. Among those discharges with CA, those who underwent R/HYST or L/HYST had fewer chronic conditions compared with O/HYST (P < 0.013 and P < 0.018, respectively); by repeated pairwise chi-square testing, those with CA who underwent R/HYST consisted of a higher proportion of diagnoses of malignancy compared with L/HYST (table 3).

Univariate analysis (table 4) identified factors increasing risk for inclusion in multivariable analysis; terms with P less than 0.2 were considered significant for inclusion in the final models (table 5). For RP, being African American conferred lower risk (OR 0.06, 95% confidence interval [CI] 0.01 to 0.44, P < 0.005) while robotic use increased risk (OR 1.55, 95% CI 1.01 to 2.36, P < 0.043).

Robotic usage increased CA risk in hysterectomy (OR 7.78, 95% CI 5.23 to 11.57, P < 0.0001), as did laparoscopy (OR 3.69, 95% CI 2.52 to 5.41, P < 0.0001). Increasing age, chronic conditions, calendar year of surgery, and a malignant diagnosis all increased CA risk, while being African American significantly lowered the risk of CA (table 4).

By multiple regression analysis that considered for prostatectomy the background covariates surgical procedure, age, number of chronic conditions, age, race, and year of surgery, robotic assistance did not significantly increase the risk of CA; for robotic assistance, OR was 1.508 (95% CI 0.987 to 2.302, P < 0.057). However, being African American was a significant negative risk factor (OR 0.062, 95% CI 0.009 to 0.446, P < 0.006; table 5). Hosmer and Lemeshow P was 0.9435 and area under the ROC curve 0.6003.

In hysterectomy, malignancy or benign diagnosis was included in the multiple regression analysis, in addition to the above, same covariates as for prostatectomy. Both laparoscopic and robotic usage (assessed as two levels in the factor variable “Modified Hysterectomy”) significantly increased CA; for L/HYST, OR was 3.821 (95% CI 2.594 to 5.630, P < 0.0001), and in R/HYST, OR was 6.505 (95% CI 4.323 to 9.788, P < 0.0001; table 5). Additional factors that increased risk in hysterectomy included advancing age (OR 1.020, 95% CI 1.007 to 1.034, P < 0.003) and increasing number of chronic conditions (OR 1.139, 95% CI 1.065 to 1.219, P < 0.0001; table 5). Hosmer and Lemeshow P was 0.6558 and area under the ROC curve 0.7850. As in prostatectomy, race was a significantly negative risk factor for CA (table 5).

Results of forward and backward elimination in the models were identical. To further test the model stringency, we tested collinearity and correlation to examine interactional effects within RP and hysterectomy. The only significant interactions were malignancy and age, and modified hysterectomy and number of chronic conditions, for hysterectomy (Supplemental Digital Content, table 1, https://links.lww.com/ALN/B140).

Elucidating the mechanisms of CA was beyond the scope of this study; however, several proposed etiologies of perioperative CA include exposure, reduced corneal hydration, and chemical or direct mechanical trauma.¹¹  Exposed corneas during general anesthesia had a significantly higher abrasion rate versus eyes that were covered, and a longer duration of operating time, that is, greater corneal exposure, increased the rate, with the peak incidence occurring between 90 and 150 min.¹²  The latter study, however, was performed approximately 40 yr ago, when risks of CA were not as well recognized, and eye protection methods were not as universally practiced as they are today, which has resulted in significant reduction in incidence of these injuries.¹³ 

Another potential mechanism of injury in RP and hysterectomy is the increase in intraocular pressure (IOP) in steep Trendelenburg position during R/PROST,¹⁴  L/HYST, and R/HYST. Surgeons request such positioning for surgical exposure and optimal robotic arm positioning. But high head down angles (40° to 45°) may not actually be necessary.¹⁵  Corneal or conjunctival edema may also occur from increased central venous pressure¹⁶  and raised IOP, causing further stress to the eye via direct fluid pressure on the globe or pressure causing the eyes to tend to remain open. Further support for these hypotheses is that IOP increased when patients were positioned prone for spine surgery, which has also been associated with a higher incidence of CAs.¹⁷ 

About 80% of RPs have robotic assistance.¹⁸  As a result of rapid introduction of the robot, there were few cases in NIS of laparoscopy alone (fig. 1). Hence, our comparison was exclusive of OP to robotic-assisted laparoscopic surgery. Accordingly, the relative contributions in RP of laparoscopy and the robot to CA risk cannot be determined using this database. Surgery in the presence of malignancy would be expected to be more technically difficult, but in our study, there was no impact upon the risk of CA in patients undergoing R/HYST or L/HYST.

An unexpected finding was the apparent protective effect against CA, in African Americans in both RP and hysterectomy. There may be previously unrecognized differences in anatomy or eye structure that render African Americans less likely to sustain CAs. Similar findings have been suggested elsewhere.¹⁹  Xian et al.,²⁰  for example, noted that African Americans had lower mortality than whites. Also, African Americans greater than 65 yr had lower adjusted 30-day mortality than whites after congestive heart failure, acute myocardial infarction, hip fracture, and gastrointestinal bleeding, which suggests a systemic versus disease-specific effect.^21,22  Our finding may warrant further study using methodology not available in the NIS.

Multivariable analysis indicated a significantly greater risk of CA with advancing age in those undergoing modified hysterectomy. While this could be explainable as a decrease in tear production with aging, it is surprising that the findings were not present for R/PROST.

There are study limitations. NIS depends on the accuracy of diagnoses and procedure codes. But the coding for hysterectomy and prostatectomy is straightforward, as is the diagnosis code for CA; hence, errors in these codes are likely not a significant factor. NIS does not provide anesthetic techniques or length of surgery. With no surgeon identification, the effect of surgeon volume and/or the learning curve cannot be assessed. Verification of diagnosis in each case is also not possible. Although diagnostic codes on discharge records are taken to reflect those acquired during hospitalization, it is also possible that some of them are preexisting conditions, in which case, the rates might be overestimated.⁹  However, CA is typically an injury from which there usually is complete recovery, and it is not likely the diagnosis code would be previously present on a subsequent discharge record.

As NIS lacks longitudinal data beyond the hospital discharge, we cannot assess the recovery from CA, its long-term impact, or increased medical costs. It would have been interesting to assess the impact of obesity; however, in less than one-fourth of the L/HYST and RP cases in 2009–2011 did we find specific body mass index codes. An important limitation of NIS in cancer studies is the lack of adjustment for tumor stage, clinical features, and pathological findings.¹⁸  Therefore, the relationship between these factors and difficulty, length, and other operative features, and CA, cannot be evaluated.

In conclusion, in a large United States nationwide sample, rates of CA have increased from 2000 to 2011 for RP and hysterectomy. In 2009–2011, there was an approximately four times higher risk when laparoscopy was used for hysterectomy, and a seven times higher risk when L/HYST was robotically assisted, compared with an open procedure. Thus, both laparoscopy and robotic assistance appear to contribute independently to increasing the risk of CA for hysterectomy. Age, the number of chronic conditions, and race were factors influencing the risk of CA. In light of the widening indications and use of robotic assistance,²³  clinicians should be vigilant and methods should be developed to lower the incidence of these eye injuries in the perioperative setting, and efforts to identify causative factors should be undertaken.

At the time this study was conducted, Dr. Sampat was a medical student at the Pritzker School of Medicine, The University of Chicago, Chicago, Illinois. Dr. Sampat was the recipient of the Calvin Fentress Medical Student Research Scholarship (Chicago, Illinois), which provided partial funding for the study. Additional funding was provided by departmental sources and by National Institutes of Health (Bethesda, Maryland) Grant RO1 EY10343 to Dr. Roth, and by a grant from the National Institutes of Health to the University of Chicago Institute for Translational Medicine (UL1 RR024999).

Dr. Roth has served as an expert witness in cases of perioperative eye injuries on behalf of patients, physicians, and hospitals. The remaining authors declare no competing interests.