Introduction

Glaucoma is the leading cause of irreversible blindness worldwide,1 and the estimated number of people affected by glaucoma will reach nearly 112 million by the year 2040, having increased from 76 million in 2020.2–4 Since the epidemiological features of glaucoma are incompletely understood, and treatment strategies are constantly being improved,5–10 it is important to assess clinical features, risk factors and treatment effectiveness of glaucoma in the population under real-world conditions.

Many observational studies using secondary data sources [eg, electronic medical records (EMRs), healthcare data, health insurance claims data or administrative data] have been published to address the clinical questions related to glaucoma, and their findings may improve disease understanding and promote further decisions in clinical practice. For example, based on California Medicare data, Yao et al found that myopia was associated with greater risk of POAG.11 Shao et al, using Taiwan’s EMR data, indicated that use of sodium glucose co-transporter 2 inhibitors was associated with a lower risk of incident glaucoma.12 Since most secondary database studies identify glaucoma by using the relevant disease diagnostic codes, the impact of misclassification bias may threaten the study validity if the coding accuracy is unclear in these secondary databases.13

So far, only three validation studies conducted in the USA and UK have assessed the accuracy of glaucoma diagnostic codes in secondary databases.14–16 However, there is still a need for more extensive evaluations of coding validity for glaucoma. For instance, none of the prior studies examined the accuracy of diagnostic codes specifically related to primary angle-closure glaucoma (PACG), a significant subtype of glaucoma prevalent in Asia.17,18 Moreover, previous validation investigations solely concentrated on the precision of ICD-9-CM codes for glaucoma,14–16 leaving the coding accuracy of ICD-10-CM glaucoma codes uncertain. To fill these research gaps, we acquired EMRs from a randomly selected group of patients who had undergone visual field examinations – considered the gold standard for diagnosing glaucoma – within Taiwan’s largest healthcare system. We proceeded to evaluate the accuracy of ICD-9-CM and ICD-10-CM codes in identifying patients with glaucoma, as well as two specific subtypes of glaucoma (primary angle-closure glaucoma, PACG and primary open-angle glaucoma, POAG).

Materials and Methods Study Settings

This study analyzed data from the Chang Gung Research Database (CGRD) which collects the EMRs from nine hospitals affiliated with the Chang Gung Medical Foundation (CGMF), the largest multi-institutional healthcare system in Taiwan.19 These CGMF hospitals, including Taipei, Keelung, Tucheng, Linkou, Taoyuan, Yunlin, Chiayi, Kaohsiung, and Fengshan branches, cover more than 10% of the total healthcare volume in Taiwan.19 As an important data source, the CGRD has provided much real-world evidence for clinical practitioners or policy makers.20–24 The study was approved by the Institutional Review Board of CGMF (IRB No: 202200878B0) and was conducted in accordance with the principles outlined in the Declaration of Helsinki. Due to its retrospective design, the requirement for informed consent was waived, ensuring patient privacy and confidentiality were upheld throughout the study.

Study Cohort

This study used the claims data reported to the National Health Insurance Administration, retrieved from the CGRD. Since visual field examinations are commonly used in clinical practice to assess suspected cases of glaucoma and exclude other diseases in the field of neuro-ophthalmology,25 we initially identified all patients who underwent visual field examination between 2011 and 2020 from the CGRD as the original study cohort. If patients received multiple visual field examinations during the study period, we only included the first result for each patient. From the original study cohort, a 2% patient sample was selected using simple random sampling, ensuring that all patients had an equal chance of being selected and no patients were selected more than once, to review their EMRs in order to assess the validity of ICD-9-CM or ICD-10-CM glaucoma codes in the claims data.

Ascertainment of Glaucoma

Two ophthalmologists (PTL and THT) independently reviewed the EMRs of the random sample taken from the original cohort,26,27 whereby any discrepancies between the reviewers were resolved through comprehensive discussion with a senior glaucoma specialist (LHC). We followed the practice guidelines from Biggerstaff et al to confirm true glaucoma cases based on patients’ reports of visual field examinations (gold standard),14 and the primary diagnostic criteria for glaucoma, including at least one of the following: 1) Cup to disc ratio of ≥ 0.6 in at least one eye. 2) Cup to disc ratio difference of ≥ 0.2 between the right and left eyes. 3) Humphrey visual field (HVF) results indicating patterns consistent with glaucomatous defects or 4) History of previous glaucoma surgery, specifically involving trabeculectomy or tube shunt implantation. Due to the clear diagnostic criteria for glaucoma, the rate of agreement on glaucoma case ascertainment between the two reviewers was 96% in this validation study. To further verify the glaucoma subtypes, we also examined gonioscopy records to determine if the angle was open or closed. For patients without gonioscopy records, the angle findings were determined through slit lamp bio-microscopy using the Van Herick method.28 Patients were classified as having POAG if gonioscopy or slit lamp bio-microscopy indicated an open angle. Conversely, patients were categorized as having PACG if an angle closure was observed.

Statistical Analysis

We evaluated the positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity with 95% confidence intervals (CI) of ICD-9-CM codes of 365.1X (open-angle glaucoma), 365.2X (angle-closure glaucoma) and ICD-10-CM codes of H4010X (unspecified open-angle glaucoma), H4011X (primary open-angle glaucoma), H4012X (low-tension glaucoma), H4020X (unspecified primary angle-closure glaucoma), H4021X (acute angle-closure glaucoma), H4022X (chronic angle-closure glaucoma), H4023X (intermittent angle-closure glaucoma), H4024X (residual stage of angle-closure glaucoma) to identify glaucoma in the CGRD.29 We also reported the coding accuracy for POAG (ICD-9-CM code: 365.1X; ICD-10-CM codes: H4010X, H4011X and H4012X), PACG (ICD-9-CM code: 365.2X; ICD-10-CM codes: H4020X, H4021X, H4022X, H4023X and H4024X) and unspecified glaucoma (ICD-9-CM code: 365.9; ICD-10-CM code: H409). Specifically, the unspecified glaucoma codes lacked the diagnostic specificity needed to identify false negative cases, making it only possible to calculate the PPV. Finally, we reported the number and frequency to summarize the reasons for false-positive glaucoma cases identified by the ICD-9-CM and ICD-10-CM codes. Data analyses were performed using SPSS version 23 (SPSS Inc., Chicago IL).

Results

We identified a total of 41,050 patients who underwent visual field examinations in the study, of which a random sample of 821 patients (2% of the total) was included for the detailed review of their EMRs (Figure 1). The average age of this random sample was 56.9 years, with 415 (50.5%) being female. A total of 390 patients out of the random sample (390/821=47.5%) were assigned ICD-9-CM or ICD-9-CM glaucoma codes. We found 359 of these (359/390=92.1%) were confirmed as true-positive glaucoma cases, and 31 (31/390=7.9%) were judged as false-positive cases identified by ICD-9-CM or ICD-10-CM glaucoma codes. A total of 431 patients out of the random sample (431/821=52.5%) were not assigned ICD-9-CM or ICD-10-CM glaucoma codes. We found 386 of these (386/431=89.6%) were judged as true-negative glaucoma cases, but 45(45/431=10.4%) were confirmed as false-negative cases identified by ICD-9-CM or ICD-10-CM glaucoma codes. Overall, the PPV, NPV, sensitivity and specificity of ICD-9-CM or ICD-10-CM codes to identify glaucoma was 92.1%, 90.0%, 88.9% and 92.6%, respectively.

Figure 1 Process of validation study.

The coding validity for each ICD-9-CM and ICD-10-CM glaucoma code is separately summarized in Table 1. The ICD-9-CM glaucoma codes demonstrated a sensitivity of 86.5%, specificity of 96.5%, PPV of 91.9%, and NPV of 90.9%, and the ICD-10-CM glaucoma codes showed a sensitivity of 87.0%, specificity of 92.8%, PPV of 92.22%, and NPV of 87.9%. The ICD-9-CM POAG codes showed a sensitivity of 64.4%, specificity of 96.5%, PPV of 91.9%, and NPV of 90.9%, and ICD-10-CM POAG codes showed a sensitivity of 57.4%, specificity of 97.4%, PPV of 92.5%, and NPV of 80.1%. In addition, the PPVs of the ICD-9-CM (91.1%) and ICD-10-CM (90.5%) low-tension glaucoma codes were similar to those of the overall coding validity for POAG. The ICD-9-CM PACG glaucoma codes revealed a sensitivity of 96.3%, specificity of 99.0%, PPV of 92.9%, and NPV of 99.5%, and ICD-10-CM PACG codes revealed a sensitivity of 73.9%, specificity of 99.4%, PPV of 89.5%, and NPV of 98.2%. The ICD-9-CM and ICD-10-CM unspecified glaucoma codes each showed PPVs of 82.1% and 81.0%, respectively.

Table 1 Accuracy of ICD-9-CM and ICD-10-CM Codes to Identify Glaucoma

In total, we identified 31 cases (31/390=7.9%) falsely coded by ICD-9-CM or ICD-10-CM glaucoma codes. We found that 58.1% of these cases were tentatively diagnosed with glaucoma but later changed to normal visual field interpretation, 25.8% of these cases were judged to be borderline glaucomatous but with insignificant visual field progression during follow-up, and 16.1% of these cases were finally judged as other diseases, such as retinitis pigmentosa, optic neuritis, age-related macular degeneration and pituitary tumor (Table 2).

Table 2 Reasons for False-Positive Glaucoma Cases

Discussion

This multi-institutional cross-sectional study was the first to report the validity of ICD-9-CM and ICD-10-CM codes to identify glaucoma cases in Taiwan, and our results indicated a PPV of 92.1% and a sensitivity of 88.9% for all types of ICD-9-CM and ICD-10-CM glaucoma codes. Specifically, for POAG, the PPV was 92.9% with a sensitivity of 61.2%. For PACG, the PPV was 92.0% with a sensitivity of 89.6%. More importantly, the validities of ICD-9-CM and ICD-10-CM glaucoma codes were consistent. Based on these findings, we could conclude that the accuracy of ICD-9-CM and ICD-10-CM glaucoma codes used in routine care data in Taiwan are acceptable and reliable for identifying glaucoma cases.

The patient characteristics of the sample in this present study show that 50.5% were female with a mean age of 56.9 years. Our results are compatible with the epidemiology of glaucoma in previous studies in Asia, reinforcing the data representativeness of potential glaucoma cases recorded in the CGRD.30,31 The PPVs for ICD-9-CM and ICD-10-CM glaucoma codes (92%) in our study were higher than those reported in two validation studies from the US (65%) and UK (84%), but similar to another report from the UK (89%),14–16 and two possible reasons may explain the difference in reported PPVs of glaucoma diagnostic codes. First, different coding practices in different healthcare systems should be noted. For example, Biggerstaff KS et al validation study in the US found that 71% of glaucoma diagnostic codes were coded by ophthalmology residents, 18% by an optometrist, 3% by an attending ophthalmologist, and 0.5% by an ophthalmology fellow in the study eye center.14 However, in this present study, the glaucoma diagnostic codes were all coded by attending ophthalmologists. Given the varying coding practices within different healthcare systems, our findings reinforce the importance of individual validation study of glaucoma diagnostic codes in different healthcare databases.27 Second, different validation algorithms among studies should also be noted. For example, Kang et al validation study in the UK used a patient self-reporting questionnaire to confirm the glaucoma diagnostic codes, which may have underestimated the true glaucoma cases. However, in this present study, the glaucoma diagnostic codes were confirmed by two independent ophthalmologists based on practice guidelines from Biggerstaff et al14 and our findings suggested more reliable and more accurate validity of ICD-9-CM and ICD-10-CM glaucoma codes.

We observed generally similar validity for ICD-9-CM and ICD-10-CM codes to identify the glaucoma subtypes, but lower sensitivity of these diagnosis codes for POAG, compared to those for PACG. We found clinical ophthalmologists often use ICD-9-CM code 365.9 or ICD-10-CM codes 365.9 or H40.9 (unspecified glaucoma) for POAG diagnoses in Taiwan. The explanation for this coding practice may be that it is due to the time constraints and high workload demands placed on ophthalmologists in the outpatient clinics in hospitals. However, we did not find this issue in cases of PACG. This could be attributed to the clinical characteristics of narrow-angle conditions, which are relatively discernible and distinct, prompting clinical ophthalmologists to select the diagnostic code specifically for PACG rather than for a more general “unspecified glaucoma” diagnostic code. Under the National Health Insurance program, patients in Taiwan have full and unrestricted access to all medical care facilities.32 For example, the number of outpatient visits per year is over 9 million in CGMF hospitals, and doctors frequently see over 50 outpatients in a morning, spending only 5 min or less for each consultation.32 Our findings suggested proper training and awareness among ophthalmologists and healthcare providers about the significance of accurate coding for POAG should be reinforced to avoid misclassification of such cases in Taiwan’s healthcare database.

Our findings could serve as an essential reference for future investigators who are interested in analyzing Taiwan’s secondary healthcare database for glaucoma research. The high PPVs of ICD-9-CM and ICD-10-CM glaucoma codes ensure their accuracy to identify true glaucoma cases. In addition, the high NPVs of ICD-9-CM and ICD-10-CM glaucoma codes indicate a low probability of missing true glaucoma cases if patients were not assigned the diagnosis codes for glaucoma.

This study has several notable strengths. First, it was the first study to validate ICD-10-CM glaucoma codes, and our results indicated that the validity of ICD-10-CM glaucoma codes is similar to that of ICD-9-CM glaucoma codes in Taiwan. Also, similar PPVs for low-tension glaucoma diagnosis codes reaffirmed our result robustness. Second, it was also the first study to validate the ICD-10-CM PACG codes, a significant subtype of glaucoma that is prevalent in Asian countries.17,18 Third, our study analyzed Taiwan’s largest multi-institutional EMR database, and our findings may be more generalizable to routine practice in Taiwan. However, it is also essential to acknowledge some limitations in our study. First, we reviewed the EMRs to assess validity of glaucoma codes in the claims data from only 821 patients (2%), who were selected by simple random sampling from the original cohort. However, this group was actually larger than in the two previous, similar validation studies.14,16 Second, we observed that 33.4% of confirmed glaucoma cases were coded by ICD-9-CM or ICD-10-CM unspecified glaucoma codes. This, in fact, may potentially reduce the sensitivity of ICD-9-CM or ICD-10-CM POAG codes. Third, this represents the largest multi-institutional validation study of glaucoma diagnostic codes in Taiwan. The CGMF hospitals account for over 10% of the total healthcare volume in Taiwan, but they do not include local clinics. Consequently, it remains uncertain whether our study results can be extrapolated to the general population served by clinics in Taiwan. Therefore, additional validation studies conducted at local clinics are necessary to corroborate our findings.

Conclusion

This study established optimal validity for the ICD-9-CM and ICD-10-CM glaucoma codes in Taiwan’s routine care data. The consistency observed in the validity of ICD-9-CM and ICD-10-CM codes indicated their accuracy for identifying glaucoma within routine care data. Our findings provide a valuable foundation for future research utilizing healthcare databases, which could contribute to enhancing the comprehension and management of glaucoma.

Author Contributions

All authors made a significant contribution to the reported work, whether in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas. All authors took part in drafting, revising or critically reviewing the article; they all gave approval for the final version to be published and all have agreed on the journal to which the article is to be submitted. Furthermore, all authors agree to be accountable for all aspects of the work.

Funding

This study was supported in part by grants from Chang Gung Medical Foundation, Taoyuan, Taiwan (CMRPG2M0012).

Disclosure

The authors report no conflicts of interest in this work.

References

1. Steinmetz JD, Bourne RRA, Briant PS. Causes of blindness and vision impairment in. 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: the right to sight: an analysis for the global burden of disease study. Lancet Glob Health. 2021;9(2):e144–e60. doi:10.1016/S2214-109X(20)30489-7

2. Karen Allison DP, Alabi O. Epidemiology of glaucoma: the past, present, and predictions for the future. Cureus. 2020;12:11.

3. Kang JM, Tanna AP. Glaucoma. Med Clin North Am. 2021;105(3):493–510. doi:10.1016/j.mcna.2021.01.004

4. Tham YC, Li X, Wong TY, et al. Global prevalence of glaucoma and projections of glaucoma burden through 2040. Ophthalmology. 2014;14:121. doi:10.1186/1471-2415-14-121

5. Wagner IV, Stewart MW, Dorairaj SK. Updates on the diagnosis and management of glaucoma. Mayo Clin Proc Innov Qual Outcomes. 2022;6(6):618–635. doi:10.1016/j.mayocpiqo.2022.09.007

6. Conlon R, Saheb H, Ahmed II. Glaucoma treatment trends: a review. Can J Ophthalmol. 2017;52(1):114–124. doi:10.1016/j.jcjo.2016.07.013

7. Harasymowycz P, Birt C, Gooi P, et al. Medical management of glaucoma in the 21st century from a Canadian perspective. J Ophthalmol. 2016;2016:6509809. doi:10.1155/2016/6509809

8. Singh K, Shrivastava A. Medical management of glaucoma: principles and practice. Indian J Ophthalmol. 2011;59 Suppl(Suppl1):S88–92.

9. Arora KS, Robin AL, Corcoran KJ, et al. Use of various glaucoma surgeries and procedures in medicare beneficiaries from 1994 to 2012. Ophthalmology. 2015;122(8):1615–1624. doi:10.1016/j.ophtha.2015.04.015

10. Weinreb RN, Aung T, Medeiros FA. The pathophysiology and treatment of glaucoma: a review. JAMA. 2014;311(18):1901–1911. doi:10.1001/jama.2014.3192

11. Yao M, Kitayama K, Yu F, et al. Association between myopia and primary open-angle glaucoma by race and ethnicity in older adults in the California medicare population. JAMA Ophthalmol. 2023;141(6):525–532. doi:10.1001/jamaophthalmol.2023.1007

12. Shao SC, Su YC, Lai EC, et al. Association between sodium glucose co-transporter 2 inhibitors and incident glaucoma in patients with type 2 diabetes: a multi-institutional cohort study in Taiwan. Diabetes Metab. 2022;48(1):101318. doi:10.1016/j.diabet.2022.101318

13. Ehrenstein V, Petersen I, Smeeth L, et al. Helping everyone do better: a call for validation studies of routinely recorded health data. Clin Epidemiol. 2016;8:49–51. doi:10.2147/CLEP.S104448

14. Biggerstaff KS, Frankfort BJ, Orengo-Nania S, et al. Validity of code based algorithms to identify primary open angle glaucoma (POAG) in Veterans Affairs (VA) administrative databases. Ophthalmic Epidemiol. 2018;25(2):162–168. doi:10.1080/09286586.2017.1378688

15. Kang EM, Pinheiro SP, Hammad TA, Abou-Ali A. Evaluating the validity of clinical codes to identify cataract and glaucoma in the UK clinical practice research datalink. Pharmacoepidemiol Drug Saf. 2015;24(1):38–44. doi:10.1002/pds.3726

16. Huerta C, Garcia Rodriguez LA, Moller CS, Arellano FM. The risk of obstructive airways disease in a glaucoma population. Pharmacoepidemiol Drug Saf. 2001;10(2):157–163. doi:10.1002/pds.567

17. Kuo CY, Ko YC, Kuang TM, Chiou PM, Chiou SH, Liu CJ. Prevalence of glaucoma in the elderly population in Taiwan: The Shihpai Eye Study. J Chin Med Assoc. 2020;83(9) :880–884.

18. Chan EW, Li X, Tham YC, et al. Glaucoma in Asia: regional prevalence variations and future projections. Br J Ophthalmol. 2016;100(1):78–85. doi:10.1136/bjophthalmol-2014-306102

19. Shao SC, Chan YY, Kao Yang YH, et al. The chang gung research database-A multi-institutional electronic medical records database for real-world epidemiological studies in Taiwan. Pharmacoepidemiol Drug Saf. 2019;28(5):593–600. doi:10.1002/pds.4713

20. Su YC, Hung JH, Chang KC, et al. Comparison of sodium-glucose cotransporter 2 inhibitors vs glucagonlike peptide-1 receptor agonists and incidence of dry eye disease in patients with type 2 diabetes in Taiwan. JAMA Network Open. 2022;5(9):e2232584. doi:10.1001/jamanetworkopen.2022.32584

21. Shao SC, Chang KC, Lin SJ, et al. Differences in outcomes of hospitalizations for heart failure after SGLT2 inhibitor treatment: effect modification by atherosclerotic cardiovascular disease. Cardiovasc Diabetol. 2021;20(1):213. doi:10.1186/s12933-021-01406-3

22. Chen HK, Shao SC, Weng MY, et al. Risk of heart failure in rheumatoid arthritis patients treated with tumor necrosis factor-alpha inhibitors. Clin Pharmacol Ther. 2021;110(6):1595–1603. doi:10.1002/cpt.2415

23. Kang EY, Shao SC, Chang KC, et al. Real-world effectiveness of intravitreal dexamethasone implants - Comparison between eyes eligible and ineligible for clinical trials and their associated outcomes. Biomed J. 2024;47(1):100607. doi:10.1016/j.bj.2023.100607

24. Su YC, Shao SC, Lai EC, et al. Risk of diabetic macular oedema with sodium-glucose cotransporter-2 inhibitors in type 2 diabetes patients: a multi-institutional cohort study in Taiwan. Diabetes Obes Metab. 2021;23(9):2067–2076. doi:10.1111/dom.14445

25. Wu Z, Medeiros FA. Recent developments in visual field testing for glaucoma. Curr Opin Ophthalmol. 2018;29(2):141–146. doi:10.1097/ICU.0000000000000461

26. Chiang MY, Shao SC, Liao SC. Validation of diagnostic codes to identify carbon monoxide poisoning in taiwan’s claims data. Front Pharmacol. 2022;13:882632. doi:10.3389/fphar.2022.882632

27. Chang C, Liao SC, Shao SC. Positive predictive values of anaphylaxis diagnosis in claims data: a multi-institutional study in Taiwan. J Med Sys. 2023;47(1):97. doi:10.1007/s10916-023-01989-2

28. Anil Kumar Srivastava VKG, Ratnendra Kumar Misra. Significance of van herick test and its comparison with gonioscopy grading. Indian J Clinical Expe Ophthalmol. 2017;3(1):11–14.

29. ICD-10 Glaucoma Reference Guide. American Glaucoma Societ American Acad Ophthalmol; 2015.

30. Zhao Y, Fu JL, Li YL, et al. Epidemiology and clinical characteristics of patients with glaucoma: an analysis of hospital data between 2003 and 2012. Indian J Ophthalmol. 2015;63(11):825–831. doi:10.4103/0301-4738.171963

31. Hong C, Hong SW, Park CK, et al. Profiles and clinical characteristics of newly diagnosed glaucoma in urban Korea: a multicenter study. Korean J Ophthalmol. 2020;34(5):353–360. doi:10.3341/kjo.2020.0033

32. Lin MH, Chang HT, Chen TJ, Hwang SJ. Why people select the outpatient clinic of medical centers: a nationwide analysis in Taiwan. PeerJ. 2020;8:e9829. doi:10.7717/peerj.9829