Real-World Outcomes of Glaucoma Filtration Surgery Using Electronic Health Records: An Informatics Study : Journal of Glaucoma

Secondary Logo

Journal Logo

New Understandings of Glaucoma: Original Studies

Real-World Outcomes of Glaucoma Filtration Surgery Using Electronic Health Records: An Informatics Study

Sun, Michelle T. MBBS, PhD; Singh, Kuldev MD, MPH; Wang, Sophia Y. MD, MS

Author Information
doi: 10.1097/IJG.0000000000002122

Abstract

With the increasing implementation of electronic health records (EHR) in ophthalmology practices worldwide, there has been a growing interest in the use of automated data extraction to facilitate the study of large-scale clinical data without the need for time and labor-intensive manual review. Robust analyses of ophthalmic surgical outcomes using EHR remain limited by free-text extraction capabilities, but recent advances by our group have demonstrated high accuracy using text-processing pipeline methods for various cataract and glaucoma surgeries.1,2

Although traditionally, tube shunt surgery was reserved for those who had either failed trabeculectomy or were at high risk of failure, the Tube Versus Trabeculectomy study and Primary Tube Versus Trabeculectomy studies have supported an expanding role of tube shunts beyond refractory glaucoma.3–5 As such, changing practice patterns have been observed worldwide with increasing rates of tube shunt procedures relative to trabeculectomy.6,7 However, real-world outcomes comparing various filtration surgeries, particularly for patients undergoing primary tube shunt insertion, remain limited. We thus sought to evaluate the long-term survival of glaucoma filtration procedures using an automated pipeline for extraction of surgical outcomes from EHR.

METHODS

Data Source/Cohort Definition

We performed a retrospective observational study of all patients undergoing glaucoma filtration surgery at the Byers Eye Institute from 2009 to 2018. We identified patients from the Stanford clinical data warehouse,8 which captures data from the Stanford EHR system, an Epic-based system (Epic Systems, Verona, WI). The database contains structured data, including patient demographics, diagnoses, and procedural codes [as per the International Classification of Diseases 9th and 10th editions9 and Current Procedural Terminology (CPT)10], medication orders (mapped from RxNorm11), eye examination findings (semi-structured data), and clinical text (unstructured data).

We included all patients undergoing trabeculectomy, Ex-PRESS shunt, Baerveldt, and Ahmed tube shunt insertion, as identified by CPT codes (trabeculectomy: 66170 and 66172, tube shunt: 66179 and 66180) and a previously validated natural language processing algorithm which determines the laterality and type of glaucoma surgery from the free text operative notes.2 Previous manual validation of the natural language processing algorithm found >98% accuracy for the extraction of preoperative and postoperative information.1 If surgical laterality was indeterminate, the surgery was excluded. Surgeries without at least 1 preoperative visit with documented intraocular pressure of the surgical eye were excluded.

Main Outcome Measures

Surgical failure was defined as cases requiring additional glaucoma surgery in the operative eye or an increase in glaucoma medication compared with that preoperatively.

Predictors

Patient characteristics were identified from EHR, including age, gender, race/ethnicity, glaucoma diagnosis, baseline IOP by applanation, baseline number of glaucoma medications used, and previous number of conjunctival-based glaucoma surgeries based on prior CPT codes. Patients’ glaucoma-related diagnoses were determined by ICD9 and ICD10 billing codes associated with the surgical encounter or clinical encounters on or before surgery. If there was more than 1 glaucoma-related diagnosis, each was prioritized in the following order: narrow or closed angle (365.2–, 365.02, 365.06, H402–, H400.3, and H400.6), secondary glaucoma (365.03, 365.3–365.6–, 365.81–3, 365.13–365.14, H400.4, H404–H406–, H408.1–H408.3–, H401.3, H401.4, and Q150), primary open angle glaucoma (365.10–365.12, 365.15, 365.89, 365.9, H401.0–H401.2–, and H401.5–), other (all other glaucoma codes). A previously validated natural language processing pipeline2 identified the laterality of glaucoma drops from medication signatures.

Statistical Analyses

Descriptive statistics were performed using the mean and SD or median and interquartile range for continuous measures, as appropriate for the distribution, with counts and percentages employed for categorical measures. Intraocular pressure and medication usage were compared between surgical category groups using 1-way ANOVA tests. Surgical failure according to the type of surgery, gender, baseline medications and a number of previous glaucoma surgeries were compared graphically using Kaplan-Meier methods and statistically using log-rank tests.

Multivariable Cox proportional hazards regression models evaluated factors associated with surgery failure. Covariates included in the models were chosen a priori and included age, gender, ethnicity, type of glaucoma surgery, glaucoma subtype, number of previous glaucoma surgeries, and number of glaucoma medications at baseline. As a sensitivity analysis, propensity score regression techniques were also utilized to evaluate the association of surgery type with the outcome. The probability of surgery type was determined using a multivariate logistic regression model, including all covariates. The propensity score was subsequently entered as a continuous variable in a Cox proportional hazards regression model. Standard errors were adjusted for patient-level clustering. The proportional hazards assumption for models was evaluated with Schoenfeld residuals and not found to be violated. Finally, predictors of intraocular pressure at various time points were assessed using linear regression. Statistical analyses were conducted using Stata Version 16, and P-value less than 0.05 was considered statistically significant.

The Stanford Institutional Review Board approved this study.

RESULTS

From 2009 to 2018, we identified 512 patients undergoing 711 consecutive glaucoma filtration surgeries performed by 5 glaucoma surgeons at a single institution (Fig. 1). Of these, there were 287 trabeculectomies, 47 Ex-PRESS shunts, 274 Baerveldt, and 103 Ahmed tube implantations. Median follow-up was 359 days (IQR 162–869) for all cases, 365 days (IQR 139–1007) for trabeculectomies, 372 days for Baerveldt (IQR 183–785), 296 days for Ahmed (IQR 146–737) and 522 days (IQR 189–832) for Ex-press shunts. The mean baseline IOP was 24.4 mm Hg (SD 10.9), and 73.1% were on ≥3 glaucoma medications preoperatively. Of all cases, 85% were primary procedures (605/711), of which 91.3% of trabeculectomies (343/424) and 80.9% of tube shunts (262/287) were primary surgeries. Table 1 describes the baseline characteristics of study subjects.

F1
FIGURE 1:
Flow chart of glaucoma surgeries included. IOP indicates Intraocular pressure.
TABLE 1 - Baseline Characteristics of Patients Undergoing Glaucoma Filtration Surgeries
Characteristic Total Population Tube Shunt Trabeculectomy
Average Age 64.8±18.7 63.2±19.7 67.2±17.0
Male Gender (%) 408 (57.4) 251 (59.2) 157 (54.7)
IOP, mmHg 24.4±10.9 27.4±10.9 20.1±9.3
Ethnicity
 Caucasian 250 (35.2) 119 (28.1) 131 (45.6)
 African American 55 (7.7) 37 (8.7) 18 (6.3)
 Asian 176 (24.8) 121 (28.5) 55 (19.2)
 Hispanic 133 (18.7) 100 (23.6) 33 (11.5)
 Unknown/Other 97 (13.6) 47 (11.1) 50 (17.4)
Glaucoma diagnosis
 POAG 372 (52.3) 168 (39.6) 204 (71.1)
 PAC 92 (12.9) 70 (16.5) 22 (7.7)
 Secondary Glaucoma 152 (21.4) 124 (29.3) 28 (9.7)
 Other 95 (13.4) 62 (14.6) 33 (11.5)
Surgery number
 First 605 (85.0) 343 (80.9) 262 (91.3)
 Second 78 (11.0) 58 (13.7) 20 (7.0)
 Third 28 (3.9) 23 (5.4) 5 (1.7)
Baseline medications
 <3 191 (26.9) 86 (20.3) 105 (36.6)
 3 188 (26.4) 120 (28.3) 68 (23.7)
 >3 332 (46.7) 218 (51.4) 114 (39.7)

Table 2 summarizes the changes in IOP and medication use over time for all patients in each surgery group. Baseline IOP and medication use were the highest in the Ahmed group (mean IOP 34.02 mm Hg, range 14 to 68 mm Hg and 3.5 medications) followed by Baerveldt (mean IOP 25.52 mm Hg, range 10 to 58 mm Hg and 3.4 medications), Ex-PRESS shunt (mean IOP 23.64 mm Hg, range 9 to 48 mm Hg and 3.2 medications) and trabeculectomy patients (mean IOP 20.06 mm Hg, range 9 to 58 mm Hg and 2.8 medications; P<0.001 for both). The mean IOP decreased in each group postoperatively, and IOP reduction was largely sustained over time in all surgical groups (Fig. 2). One year postoperatively, tube shunt patients had a 2.53 mm Hg (95% CI 0.92 to 4.15, P=0.002) higher IOP as compared with trabeculectomy patients when adjusted for age, gender, ethnicity, type of glaucoma surgery, glaucoma subtype, number of previous glaucoma surgeries and number of baseline glaucoma medications. This difference in IOP remained significant 2 years postoperatively (2.24 mm Hg greater in tube shunts, 95% CI 0.65 to 3.83, P=0.006) but was attenuated and no longer statistically significant by 3 years (1.50 mm Hg, 95% CI −0.62 to 3.62, P=0.17).

TABLE 2 - IOP and Medication Use Following Glaucoma Filtration Surgery
Trabeculectomy Ex-PRESS Baerveldt Ahmed P
Baseline
 IOP (mm Hg) 20.1±9.3 23.6±9.8 25.5±9.1 34.0±12.7 <0.001
 Medications N=711 2.8±1.6 3.2±1.3 3.4±1.3 3.5±1.3 <0.001
1 Mo
 IOP (mm Hg) 10.5±5.7 11.5±6.1 17.7±8.8 19.3±8.5 <0.001
 Medications N=662 1.9±1.7 1.3±1.6 3.1±1.4 2.2±1.9 <0.001
3 Mo
 IOP (mm Hg) 10.3±5.1 11.9±5.4 15.1±5.8 17.7±8.7 <0.001
 Medications N=577 1.3±1.6 0.8±1.3 2.4±1.6 2.5±1.7 <0.001
1 Y
 IOP (mm Hg) 11.1±6.0 14.9±6.3 13.3±4.9 16.6±6.1 <0.001
 Medications N=325 1.2±1.5 1.5±1.5 2.5±1.7 2.6±1.3 <0.001
2 Y
 IOP (mm Hg) 11.2±4.4 16.4±7.7 13.9±4.6 15.9±5.9 <0.001
 Medications N=203 1.4±1.7 1.7±1.5 2.3±1.5 2.5±1.5 <0.001
3 Y
 IOP (mm Hg) 12.0±7.3 12.8±5.3 14.6±6.6 15.6±6.7 0.043
 Medications N=186 1.3±1.6 1.7±1.6 2.2±1.6 2.7±1.9 <0.001
4 Y
 IOP (mm Hg) 11.7±5.7 13.5±2.7 15.2±6.9 18.6±10.3 0.001
 Medications N=132 1.8±1.9 3.2±1.7 2.2±1.6 2.7±2.1 0.102
5 Y
 IOP (mm Hg) 12.6±6.2 16.0±3.6 16.2±7.2 17.3±9.3 0.096
 Medications N=85 2.1±1.8 2.0±0.8 2.6±1.7 2.1±1.5 0.504

F2
FIGURE 2:
Change in intraocular pressure over time following glaucoma surgery. Figure 1 demonstrates the mean change in intraocular pressure at baseline before glaucoma filtration surgery and then at postoperative months 1, 3, 12, 24, 36, 48 and 60 stratified by type of surgery. Bands indicate 95% confidence intervals.

We compared surgical failure according to various baseline characteristics. Kaplan-Meier plots are shown in Figure 3. There was a statistically significant difference in surgical failure between the different types of glaucoma filtration procedures (P=0.0128), sex (P=0.028), baseline glaucoma medication (P<0.001) and a number of previous surgeries (P<0.001). At 1 year, the failure rate for trabeculectomy was 16.6%, as compared with 25.9% for Ex-PRESS shunt implantation, 23.9% for Baerveldt implantation, and 31.0% for Ahmed tube implantation. At 3 years, failure rates were 30.7% for trabeculectomy, 48.3% for Ex-PRESS shunt, 48.2% for Baerveldt, and 48.3% for Ahmed implantation. By 5 years, cumulative failure was 45.5% for trabeculectomies, 54.4% for Ex-PRESS shunts, 59.6% for Baerveldt implants, and 67.1% for Ahmed implants.

F3
FIGURE 3:
Kaplan-Meier plots demonstrating time to surgical failure according to baseline characteristics. Kaplan-Meier plots demonstrating time to surgical failure based upon surgical type (top left), sex (top right), baseline medications (bottom left) and number of previous glaucoma surgeries (bottom right).

Table 3 summarizes the results of multivariable Cox proportional hazards models comparing survival of different types of glaucoma filtration surgeries, adjusted for age, gender, ethnicity, type of glaucoma surgery, glaucoma subtype, number of previous glaucoma surgeries, and number of glaucoma medications at baseline. As compared with trabeculectomy, Baerveldt tube implantation was associated with a 1.44-fold higher risk of failure (hazard ratio (HR) 1.44, 95% CI 1.02 to 2.02), while Ahmed tubes were associated with a 2.01-fold increased risk of failure (HR 2.01, 95% CI 1.28 to 3.17). There was a trend toward a higher hazard of failure for Ahmed tubes compared with Baerveldt tubes, but this did not reach statistical significance (HR 1.40, 95% CI 0.93 t 2.10). Previous conjunctival-based glaucoma surgery was also a significant independent risk factor associated with a higher failure rate regardless of surgery type, with a 1.93-fold increased risk of failure following 1 glaucoma surgery (HR 1.93, 95% CI 1.32 to 2.83) which increased to a 2.74-fold higher risk of failure after 2 or more previous surgeries (HR 2.74, 95% CI 1.62 to 4.64). Compared with patients taking 3 or more glaucoma medications at baseline, patients taking fewer baseline medications had increased surgical failure (<3 medications: HR 2.96, 95% CI 2.12 to 4.13; 3 medications: HR 1.68, 95% CI 1.21 to 2.34). Men had increased rates of surgical failure compared with women (HR 1.40, 95% CI 1.03 to 1.90). In a propensity score-adjusted model, tube shunt surgery remained associated with an increased risk of failure as compared with trabeculectomy (HR 1.47, 95% CI 1.08 to 2.02). In supplemental analyses where only primary glaucoma surgeries were included, we found the above results largely unchanged; specifically, there remained a higher risk of failure associated with both Baerveldt and Ahmed tubes as compared with trabeculectomy (Supplement Table 1, Supplemental Digital Content 1, https://links.lww.com/IJG/A662).

TABLE 3 - Multivariable Associations of Surgical Failure
Characteristic Hazard Ratio 95% CI P
Age 1.00 0.99–1.01 0.99
Males 1.40 1.03 to 1.90 0.03
Baseline IOP, mm Hg 1.01 1.00–1.02 0.17
Ethnicity
 Caucasian Reference
 African American 0.79 0.47–1.31 0.36
 Asian 0.93 0.63–1.38 0.73
 Hispanic 1.15 0.73–1.82 0.55
 Unknown/Other 0.80 0.49–1.30 0.37
Glaucoma diagnosis
 POAG Reference
 PAC 1.03 0.67–1.57 0.91
 Secondary glaucoma 0.68 0.44–1.04 0.07
 Other 1.28 0.79–2.05 0.32
Surgery type
 Trabeculectomy Reference
 Ex-Press shunt 1.62 0.96–2.73 0.072
 Baervedlt tube 1.44 1.02 to 2.02 0.037
 Ahmed tube 2.01 1.28 to 3.17 0.002
Surgery number
 First Reference
 Second 1.93 1.32 to 2.83 0.001
 Third 2.74 1.62 to 4.64 <0.001
Baseline medications
 >3 Reference
 3 1.68 1.21 to 2.34 0.002
 <3 2.96 2.12 to 4.13 <0.001

DISCUSSION

In this study, we used an automated text-processing pipeline to extract surgical outcomes from EHR to analyze the long-term survival of various glaucoma filtration surgeries and found that tube shunt surgery was associated with a significantly higher risk of failure as compared with trabeculectomy. Previous conjunctival-based glaucoma surgery, fewer baseline medications, and male sex were also significant risk factors for surgical failure. Our findings demonstrate the utility of applying an informatics pipeline to EHR to investigate real-world outcomes for various surgical procedures with numerous potential applications.

The growing use of EHR and online registries creates a wealth of opportunities to study large-scale population-based data. However, registries often provide only cross-sectional information with limited data points available.12,13 Furthermore, EHR-leveraged studies tend to rely on structured text field extraction. For example, a recent study utilizing EHR to study trabeculectomy outcomes included data from specific clinical fields, such as IOP and best-corrected visual acuity but excluded free-text fields.14 Long-term outcomes requiring more detailed granular data for larger populations remain challenging to obtain. As such, we continue to rely on smaller retrospective reviews or trials, which have stringent inclusion criteria that may not be broadly applicable in the real world. Having developed an automated text-processing pipeline for free-text data, which performed with high accuracy for cataract and glaucoma surgery details,2 our group subsequently utilized this pipeline to evaluate long-term changes in IOP following cataract surgery in 7574 eyes of 4883 patients with and without glaucoma demonstrating the utility and feasibility of this method to expand our understanding of real-world outcomes with large patient populations.1 In the current study, being able to extract clinical narrative text allowed analysis of clinical details beyond structured data fields. This allowed the inclusion of glaucoma implant laterality and type, which is not possible to obtain from CPT codes alone due to coding overlap and medication information that formed part of our surgical failure criteria in keeping with our real-world focus.

In recent years, changing patterns of practice have since emerged following the Tube Versus Trabeculectomy (TVT) study in 2014 and subsequently the Primary Tube Versus Trabeculectomy (PTVT) trial in 2018,3,4 which supported an expanding role of tube shunts beyond refractory glaucoma.15,16 The TVT recruited patients with a history of cataract or glaucoma surgery, while the PTVT enrolled patients without any history of ocular surgery. In both studies, surgical failure was defined as IOP more than 21 mm Hg, less than 5 mm Hg or less than 20% reduction on 2 consecutive follow-up visits, re-operation, or loss of light perception vision. At the 3-year follow-up, no significant difference was found in the PTVT study, but surgical failure was significantly higher for trabeculectomy compared with tubes in the TVT study.5 This was in keeping with previous studies, which suggest that conjunctival inflammation and scarring from prior surgeries contribute to higher rates of failure in trabeculectomy.17–19 Similarly, we found that previous conjunctival-based glaucoma surgery was a significant risk factor for failure in all glaucoma surgeries. Previous studies have also postulated that preservatives in medication may precipitate increased inflammation and altered healing hence predisposing to a higher risk of failure.19,20 Although we found that being on fewer than 3 medications at baseline was associated with a higher hazard of surgical failure, this may relate to how we defined surgical failure but could also reflect a tendency to favor conservative medical therapy (ie, more medication) over surgery for patients with less severe disease. Male sex was also a significant predictor of failure, which has previously been shown in both the Ahmed and Baerveldt implants.21 While not found to be a significant predictor of failure in long-term trabeculectomy studies,20 male sex has been associated with a higher risk of tenon cyst formation post trabeculectomy which is known to increase the risk of needing subsequent medical or surgical intervention,22 and has also been shown to confer a higher risk of failure in patients with uveitic glaucoma undergoing trabeculectomy.23

We found that tube shunts were associated with a significantly higher risk of requiring additional medication or re-operation as compared with trabeculectomy in our study of predominantly primary glaucoma surgeries (80.9% of tube shunts and 91.3% of trabeculectomies were primary). We also found a statistically significant lower IOP in those undergoing trabeculectomy as compared with tube shunts at 2 years postoperatively, although this difference was attenuated and no longer significant by 3 years onwards. Our results support existing evidence which suggests that trabeculectomy may yield lower postoperative IOP, particularly in the initial postoperative period and thus impact postoperative medication status. The PTVT study found that trabeculectomy achieved greater IOP reduction with less need for supplemental medication postoperatively,4 and similarly, a retrospective review of primary Baerveldt versus trabeculectomy showed significantly fewer trabeculectomy patients required adjunctive medical therapy at 3 years follow-up with a trend toward lower IOP.24 However, the primary Ahmed versus trabeculectomy trial showed no difference in adjunctive medications at a mean of 31 months of follow-up, although trabeculectomy patients were found to have lower IOPs in the first year.25 Furthermore, the Ahmed versus Baerveldt study has previously demonstrated lower failure rates and lower IOP with the Baerveldt group,26 with subsequent studies also showing higher re-operation rates with Ahmed tubes.27 While we found a trend toward a higher hazard of failure for Ahmed tubes compared with Baerveldt tubes, this did not reach statistical significance. Furthermore, the higher preoperative IOP for Ahmed patients compared with Baerveldt patients likely reflected tube shunt practice patterns, favoring the valved tube for patients with higher IOP requiring more immediate IOP lowering as compared with the nonvalved Baerveldt. Additional larger studies are required to better investigate differences between the tube shunt outcomes.

Strengths of our study include the novel use of an automated text-processing pipeline capable of handling structured and unstructured fields from EHR to maximize the amount of clinical information able to be extracted for analysis. In keeping with our real-world focus, we defined failure using the practical endpoint of requiring either additional medical or surgical treatment and elected not to use IOP criteria, which differed from previous studies in the literature. This reflects the fact that the target IOP for each patient varies,28 and the same IOP or IOP reduction could be considered either a ‘success’ or ‘failure’ depending upon the individual case. We believe this enhances the applicability of our results in the day-to-day clinic setting.

Our study has a number of limitations which warrant recognition. Although we have previously demonstrated high accuracy of our automated text-processing extraction methods using manual data validation,1 as with any EHR, there is the possibility of data-entry errors, such as from outdated medication lists, previous information carried forward or miscoding.29 Within the limits of our automated pipeline, we were unable to extract additional clinical details, such as specific clinical exam findings or target IOPs which may have provided additional valuable insights. In addition, as with any observational study, a number of patients were lost to follow-up, although the rates were similar across each surgical category. Furthermore, our clinical information was extracted from a single academic centre as reflected in our smaller patient population, and the demographic of patients in our area may not be representative of the general population. In addition, it is possible that patients may have changed treating institutions following surgery, and thus their outcomes would be unknown and censored. With all real-world surgical studies, there are also inherent surgeon preferences introducing surgical bias, which should be considered when interpreting results. As we continue to advance our automated extraction capabilities, additional robust analyses capable of analyzing more detailed data will become possible. In addition to limitations relating to data extraction capabilities, there are, of course, numerous limitations when comparing different surgical procedures that are performed without randomization. For example, tube implantation is more likely to be performed than trabeculectomy in eyes with neovascular, uveitic, and other high risks for trabeculectomy failure glaucoma diagnoses, and such eyes are generally more difficult to treat, all other things being equal than eyes with primary open angle glaucoma. All of the limitations of a retrospective study design must be considered when drawing conclusions from such real-world approaches.

In summary, we have demonstrated the utility of an automated pipeline for data extraction from EHR to study long-term outcomes of glaucoma filtration surgeries. This approach can provide real-world information on the success of various glaucoma procedures in the respective populations that receive such treatment. While acknowledging the limitations of making comparisons between procedures using this retrospective approach, we have shown that Baerveldt and Ahmed tube shunt implantation was associated with increased failure rates compared with trabeculectomy. Fewer baseline glaucoma medications, more previous glaucoma surgeries, and male sex were also risk factors for surgical failure. As we continue to refine our methods for data extraction, we expect to obtain increasing granularity in our quest to answer important clinical questions pertaining to glaucoma surgical outcomes.

REFERENCES

1. Wang SY, Azad AD, Lin SC, et al. Intraocular pressure changes after cataract surgery in patients with and without glaucoma: An informatics-based approach. Ophthalmol Glaucoma. 2020;3:343–349.
2. Wang SY, Pershing S, Tran E, et al. Automated extraction of ophthalmic surgery outcomes from the electronic health record. Int J Med Inform. 2020;133:104007.
3. Gedde SJ, Schiffman JC, Feuer WJ, et al. Tube versus Trabeculectomy Study G. Treatment outcomes in the Tube Versus Trabeculectomy (TVT) study after five years of follow-up. Am J Ophthalmol. 2012;153:789–803.e2.
4. Gedde SJ, Feuer WJ, Lim KS, et al. Primary Tube Versus Trabeculectomy Study G. Treatment outcomes in the primary tube versus trabeculectomy study after 3 years of follow-up. Ophthalmology. 2020;127:333–345.
5. Gedde SJ, Feuer WJ, Chen PP, et al. Primary Tube Versus Trabeculectomy Study G. Comparing treatment outcomes from the tube versus trabeculectomy and primary tube versus trabeculectomy studies. Ophthalmology. 2021;128:324–326.
6. Rathi S, Andrews CA, Greenfield DS, et al. Trends in glaucoma surgeries performed by glaucoma subspecialists versus nonsubspecialists on Medicare beneficiaries from 2008 through 2016. Ophthalmology. 2021;128:30–38.
7. Sun MT, Madike R, Huang S, et al. Changing trends in glaucoma surgery within Australia. Br J Ophthalmol. 2022;106:957–961.
8. Lowe HJ, Ferris TA, Hernandez PM, et al. STRIDE--An integrated standards-based translational research informatics platform. AMIA Annu Symp Proc. 2009;2009:391–395.
9. World Health Organization. International Classification of Diseases. 2016. Available at: http://www.who.int/classifications/icd/en/. Accessed December 14, 2017.
10. American Medical Association. CPT® (Current Procedural Terminology). In: American Medical Association. CPT® (Current Procedural Terminology). 2017. Available at: https://www.ama-assn.org/practice-management/cpt-current-procedural-terminology. Accessed December 14, 2017.
11. RxNorm. META2017AB Full Update 2018_03_05. Bethesda, MD: National Library of Medicine.
12. Zetterstrom C, Behndig A, Kugelberg M, et al. Changes in intraocular pressure after cataract surgery: analysis of the Swedish National Cataract Register Data. J Cataract Refract Surg. 2015;41:1725–1729.
13. Leal I, Chu CJ, Yang YY, et al. Intraocular pressure reduction after real-world cataract surgery. J Glaucoma. 2020;29:689–693.
14. Guo X, Li B, Friedman DS, et al. Surgical outcomes and quality assessment of trabeculectomy: Leveraging electronic health records for clinical data visualization. J Glaucoma. 2019;28:1023–1028.
15. Gedde SJ, Feuer WJ, Chen PP, et al. Primary Tube Versus Trabeculectomy Study G. Comparing treatment outcomes from the Tube Versus Trabeculectomy and Primary Tube Versus Trabeculectomy Studies. Ophthalmology. 2021;128:324–326.
16. Chen J, Gedde SJ. New developments in tube shunt surgery. Curr Opin Ophthalmol. 2019;30:125–131.
17. Broadway DC, Grierson I, Hitchings RA. Local effects of previous conjunctival incisional surgery and the subsequent outcome of filtration surgery. Am J Ophthalmol. 1998;125:805–818.
18. Sugimoto Y, Mochizuki H, Ohkubo S, et al. Intraocular Pressure outcomes and risk factors for failure in the collaborative bleb-related infection incidence and treatment study. Ophthalmology. 2015;122:2223.
19. Issa de Fendi L, Cena de Oliveira T, Bigheti Pereira C, et al. Additive effect of risk factors for trabeculectomy failure in glaucoma patients: A Risk-group from a cohort study. J Glaucoma. 2016;25:e879–e883.
20. Landers J, Martin K, Sarkies N, et al. A twenty-year follow-up study of trabeculectomy: risk factors and outcomes. Ophthalmology. 2012;119:694–702.
21. Goulet RJ III, Phan AD, Cantor LB, et al. Efficacy of the Ahmed S2 glaucoma valve compared with the Baerveldt 250-mm2 glaucoma implant. Ophthalmology. 2008;115:1141–1147.
22. Feldman RM, Gross RL, Spaeth GL, et al. Risk factors for the development of Tenon’s capsule cysts after trabeculectomy. Ophthalmology. 1989;96:336–341.
23. Ceballos EM, Beck AD, Lynn MJ. Trabeculectomy with antiproliferative agents in uveitic glaucoma. J Glaucoma. 2002;11:189–196.
24. Panarelli JF, Banitt MR, Gedde SJ, et al. A retrospective comparison of primary Baerveldt Implantation versus Trabeculectomy with Mitomycin C. Ophthalmology. 2016;123:789–795.
25. Wilson MR, Mendis U, Paliwal A, et al. Long-term follow-up of primary glaucoma surgery with Ahmed glaucoma valve implant versus trabeculectomy. Am J Ophthalmol. 2003;136:464–470.
26. Christakis PG, Zhang D, Budenz DL, et al. Five-year pooled data analysis of the Ahmed Baerveldt comparison study and the Ahmed Versus Baerveldt Study. Am J Ophthalmol. 2017;176:118–126.
27. Kung FF, Knier CG, Garmany A, et al. Need for additional glaucoma surgery and complications following glaucoma drainage device surgery. J Glaucoma. 2021;30:508–514.
28. Quigley HA. 21st century glaucoma care. Eye (Lond). 2019;33:254–260.
29. Hersh WR, Weiner MG, Embi PJ, et al. Caveats for the use of operational electronic health record data in comparative effectiveness research. Med Care. 2013;51:S30–S37.
Keywords:

automated data extraction; electronic health records; glaucoma surgery; health outcomes; health services; real-world data; natural language processing

Supplemental Digital Content

Copyright © 2022 Wolters Kluwer Health, Inc. All rights reserved.