Board Certification Is Associated With a Reduced Risk of Endophthalmitis After Intravitreal Injections

Abstract

Purpose:

This work investigates associations between physician qualifications and the risk of postintravitreal injection endophthalmitis.

Methods:

This retrospective analysis of data from medical claims studied Medicare beneficiaries undergoing 1 or more intravitreal injections between January 1, 2013, and December 31, 2017. Logistic regression analysis was performed to assess whether board certification status or retina subspecialty training was associated with lower risk of postinjection endophthalmitis, controlling for patient's age, race, and sex, type of agent injected, diagnosis, and year of injection. The main outcome measure was odds ratio (OR) of receiving a diagnosis of endophthalmitis in the 14 days after intravitreal injection. Clinical outcome and quality of care were not evaluated in this study.

Results:

A total of 2 907 324 intravitreal injections were performed on 219 640 patients by 4315 ophthalmologists, 3196 (74%) of whom were retina specialists and 4021 (92%) of whom were certified by the American Board of Ophthalmology (ABO). Overall, there were 1088 (0.037%) cases of postinjection endophthalmitis, of which 1024 (0.037%) were injected by ABO-certified ophthalmologists and 64 (0.050%) by non–board-certified ophthalmologists. Injections by ABO-certified ophthalmologist had 28% reduced odds of endophthalmitis (OR = 0.72; 95% CI, 0.523-0.996, P = .05). Higher odds of endophthalmitis were observed for corticosteroid injections (OR = 3.91; 95% CI, 2.75-5.56, P < .001) and aflibercept injections (OR = 1.47; 95% CI, 1.19-1.80, P < .001). Patients' sex and race, the diagnosis associated with the injection, and providers' retina subspeciality training were not associated with the rate of endophthalmitis (P < .20 for all comparisons).

Conclusions:

We found evidence that endophthalmitis may be reduced when ABO-certified physicians perform an intravitreal injection.

Introduction

Endophthalmitis is the most serious complication of intravitreal injection and often results in permanent vision loss. ¹ At least 3 large retrospective studies have shown an incidence of postinjection endophthalmitis ranging from 0.015% to 0.08%. ^{2 -4} Known and suspected factors associated with endophthalmitis include a patient's age, ⁵ female sex, ^{5 -7} corticosteroid, ^8,9 aflibercept, ⁹ and agents that are not prepackaged. ^6,7 Patients and ophthalmologists alike desire to avoid endophthalmitis after intravitreal injection because of the devastating consequences of this complication.

Meanwhile, intravitreal injection is increasing in volume by 6% per year in the United States, ¹⁰ raising the question of how to manage the growing treatment burden on ophthalmologists if current volumes increase as populations age. The United Kingdom has addressed its shortage of capacity for intravitreal injection through a program of nurse-administered intravitreal injections supervised by a vitreoretinal consultant. ¹¹ In contrast, intravitreal injections in the United States are administered by ophthalmologists (usually fellowship-trained retina specialists). ¹⁰ Because of their training, retina specialists are most prepared to manage the complications of intravitreal injection, including endophthalmitis. However, whether the training and/or certification of the provider affects the risk of endophthalmitis has yet to be evaluated. In this study we examined whether either the American Board of Ophthalmology's (ABO) board certification of the provider or fellowship training in the retina field is associated with endophthalmitis after intravitreal injection.

Methods

For this retrospective cohort study, we used medical claims data from the Centers for Medicare and Medicaid Services between 2013 and 2017. All intravitreal injections (defined as patients with a Current Procedural Terminology code of 67028) were evaluated for possible inclusion. Injections that were coded both as “LT” (left) and “RT” (right) or with a “–50” (bilateral) modifier the same day were counted as 2 injections. All patients were required to have at least 6 months of Medicare coverage prior to and 1 month after the injection to be included in this study. We excluded patients with a previous diagnosis of endophthalmitis, multiple drugs injected the same day, or intraocular surgery 14 days before or after the injection. We also excluded injections not associated with a National Provider Identifier number.

The injected drug (aflibercept, bevacizumab, ranibizumab, or corticosteroid) and the diagnosis for which the intravitreal injection was administered (age-related macular degeneration, diabetic macular edema, proliferative diabetic retinopathy, retinal vein occlusion, or other) were identified via the International Classification of Diseases, revisions 9 and 10, and drug codes billed the same day (Supplemental Table 1, available online).

ABO board-certification status was determined for each provider from ABO's “Verify a Physician” webpage ( https://abop.org/verify-a-physician; accessed November-December 2019). Postresidency retina fellowship training was determined using an internet search algorithm previously described, ¹⁰ from the webpages of the ABO (https://abop.org/verify-a-physician), American Academy of Ophthalmology (AAO; https://secure.aao.org/aao/find-ophthalmologist), American Society of Retina Specialists (ASRS; https://www.asrs.org/find-a-specialist), Doximity (https://www.doximity.com/directory/physicians), and the proceduralist’s clinic and/or hospital webpage as identified by Google (accessed October 2019-April 2020). Discrepancies on the various webpages were uncommon, but when they were encountered we considered case by case whether to favor the information listed on the ABO, AAO, or ASRS webpages.

The main outcome measure was the odds of endophthalmitis after intravitreal injection. This was defined as any incident diagnosis of endophthalmitis in conjunction with treatment of endophthalmitis (injection, “tap and inject,” or vitrectomy).

Multivariable logistic regression analysis with generalized estimating equations was performed after incorporating the intrapatient injections' dependency into the estimation to control for covariates including age, sex, race, intravitreal agent, injection-associated diagnosis, and year of injection. Statistical analyses were performed using SAS/STAT (version 15.1).

Results

A total of 2 907 324 intravitreal injections (219 640 patients) met the inclusion criteria (Figure 1). Injections were administered by 4021 providers who were ABO board certified and 330 who were not (see Table 1 for descriptive statistics). A total of 2 779 751 injections performed by ABO-certified physicians and 127 573 injections by non–ABO-certified physicians resulted in 1024 (0.037%) and 64 (0.050%) endophthalmitis outcomes, respectively. The 3196 (73.9%) providers who had completed a retina fellowship administered 2 742 505 (94.3%) of the injections, resulting in 1027 (0.04%) endophthalmitis outcomes, compared with 61 (0.04%) endophthalmitis outcomes after 164 819 (5.7%) injections by 1155 (26.5%) nonretina providers. Univariable analysis of all variables is shown in Supplemental Table 2 (available online). After multivariable logistic regression, ABO board–certified providers had 28% reduced odds of endophthalmitis (odds ratio [OR] = 0.72; 95% CI, 0.523-0.996, P = .047; see Table 2 for full multivariable model results). By contrast, retina subspeciality training was not associated with endophthalmitis (OR = 1.01; 95% CI, 0.75-1.37, P = .94).

Figure 1.

Flowchart for the selection of records in the Medicare claims database (January 1, 2013-December 31, 2017) for inclusion in and exclusion from the analysis.

Table 1.

Study Population Characteristics in the ABO Certified vs Non–ABO-Certified Injection Cohorts.^a

Variableb	All injections	ABO-certified No. of injections (%)	Non–ABO-certified No. of injections (%)	P valuec
No. of providers	4351	4021 (92.4)	330 (7.3)
No. of injections	2 907 324	2 779 751 (95.6)	127 573 (4.4)
No. of patients	219 640	211 553 (93.7)	14 287 (6.3)
Endophthalmitis outcomes, No. (%)	1088 (0.037)	1024 (0.037)	64 (0.050)	.100
Patient sex (female %)	1 791 286 (61.6)	1 713 513 (61.6)	77 755 (60.9)	.294
Patient age at injection, y, No. (%)				.0001
<65	105 156 (3.62)	99 949 (3.6)	5687 (4.5)
65-74	655 610 (22.6)	626 514 (22.5)	29 096 (22.8)
75-84	1 119 773 (38.5)	1 070 347 (38.5)	49 426 (38.7)
85-94	937 159 (32.2)	897 154 (32.3)	40 005 (31.4)
≥95	89 626 (3.08)	86 267 (3.1)	1359 (2.6)
Patient race/ethnicity, No. (%)				.0007
White	2 681 568 (92.2)	2 563 844 (92.2)	117,724 (92.3)
Black	97 734 (3.36)	93 623 (3.4)	4111 (3.2)
Hispanic	38 597 (1.33)	36 343 (1.3)	2254 (1.8)
Asian	36 966 (1.27)	35 788 (1.3)	1178 (0.9)
Native/other	52 459 (1.80)	50 153 (1.8)	2306 (1.8)
Medication injected, No. (%)				<.0001
Ranibizumab	1 154 778 (39.7)	628 099 (22.6)	26 044 (20.4)
Bevacizumab	654 143 (22.5)	1 093 677 (39.3)	61 101 (47.9)
Aflibercept	884 464 (30.4)	853 140 (30.7)	31 324 (24.6)
Corticosteroid	45 443 (1.56)	43 359 (1.6)	2084 (1.6)
Other	168 496 (5.80)	161 476 (5.8)	7020 (5.5)
Disease treated, No. (%)d				<.0001
AMD	2 237 037 (76.9)	2 140 066 (77.0)	96 971 (76)
Diabetic retinopathy	354 396 (12.2)	338 041 (12.2)	16 355 (12.8)
Vein occlusion	292 105 (10.0)	279 238 (10.1)	12 867 (10.1)
Other	23 786 (0.82)	22 406 (0.8)	1380 (1.1)
No. of injections per y, No. (%)				<.0001
2013	526 574 (18.1)	498 600 (17.9)	27 974 (21.9)
2014	552 895 (19.0)	527 331 (19.0)	25 564 (20.0)
2015	589 924 (20.3)	564 960 (20.3)	24 964 (19.6)
2016	634 846 (21.8)	610 147 (21.9)	24 699 (19.4)
2017	603 085 (20.7)	578 713 (20.8)	24 372 (19.1)

Abbreviations: ABO, American Board of Ophthalmology; AMD, age-related macular degeneration.

^aPatients' sex, age, and race are identified separately for each injection. Medication injected, disease treated, and provider are also identified for each injection, as these variables may change from injection to injection.

^bSome patients received injections with more than 1 agent, in more than 1 year, and had more than 1 condition necessitating an injection.

^c P values were calculated using the Rao-Scott χ² method after adjusting for the intrapatient injections correlation.

^dDisease treated (other) includes choroidal neovascularization, myopic degeneration, retinal edema, and other less common diagnoses, excluding intraocular infection.

Table 2.

Multivariable Logistic Regression Analysis of Probability of Endophthalmitis After Intravitreal Injection.^a

Variable	Odds ratio	95% CI	P value
Drug, vs ranibizumab
Bevacizumab	0.94	0.76-1.16	.584
Aflibercept	1.47	1.19-1.80	.0003
Corticosteroid	3.91	2.75-5.56	<.0001
Other	1.46	1.06-2.01	.020
Age, y, vs 65-74 y
<65	1.53	1.06-2.19	.022
75-84	0.94	0.77-1.15	.550
85-94	0.75	0.60-0.94	.012
≥95	0.50	0.29-0.85	.011
Sex, vs female
Male	0.99	0.85-1.15	.874
Race/ethnicity, vs White
Black	1.14	0.76-1.71	.519
Hispanic	1.19	0.63-2.23	.599
Asian	0.99	0.50-1.98	.979
Native/other/unknown	0.87	0.51-1.48	.611
Disease treated, vs AMD
Diabetic retinopathy	0.94	0.71-1.25	.686
Vein occlusion	0.86	0.67-1.09	.204
Other	1.31	0.72-2.35	.376
Injection y, vs 2013
2014	0.77	0.60-0.99	.038
2015	0.81	0.63-1.03	.084
2016	0.90	0.72-1.14	.302
2017	1.02	0.81-1.28	.883
Provider characteristics, vs none
Retina fellowship	1.00	0.73-1.36	>.99
ABO board certification	0.72	0.52-1.00	.047

Abbreviations: ABO, American Board of Ophthalmology; AMD, age-related macular degeneration.

^aAge, sex, and race refer to the patient, not the provider. Disease treated (other) includes choroidal neovascularization, myopic degeneration, retinal edema, and other less common diagnoses, excluding intraocular infection.

Regarding covariates, corticosteroids (OR = 3.91; 95% CI, 2.75-5.56, P < .001), aflibercept (OR = 1.47; 95% CI, 1.19-1.80, P < .001), and patient age younger than 65 years (OR = 1.53; 95% CI, 1.06-2.19, P = .02) were associated with higher odds of endophthalmitis. No statistically significant difference was found between bevacizumab and ranibizumab (P = .59). Patient's sex and race and injection diagnosis were not associated with endophthalmitis (P < .05 for all comparisons).

Conclusions

This series of 2 907 324 injections was the largest to date that studied endophthalmitis outcomes. The observed endophthalmitis rate of 0.04% was low and within the range previously reported. ^{1 -4,6,8,9} This study found reduced odds of endophthalmitis after intravitreal injection for ABO board–certified ophthalmologists, as compared with noncertified providers. Our results also concur with previous studies that found a higher risk of infection with corticosteroids ^8,9 and younger age. ⁵

This study investigated the association between physicians' qualifications and the risk of developing endophthalmitis after intravitreal injection. The topic is timely as health care systems face growing treatment burden in the setting of limited resources. In the United Kingdom, nurses assist the vitreoretinal service in administering intravitreal injections to aid capacity. ¹¹ In the United States, increasing the scope of practice for nonphysicians is an active and ongoing debate ¹² ; however, our findings suggest that the potential for higher odds of endophthalmitis for providers who are not ABO board–certified should raise concern about extending administration of intravitreal injection to less-credentialed providers.

Critics of board certification contend that certification and maintenance of certification are costly and time consuming for physicians. However, within the ophthalmology community, most believe that certification is an important tool in maintaining the highest standards of medical care. ¹³ In fact, the present study confirmed that most ophthalmology providers (92.4%) are board certified.

Aflibercept was found in this study to have higher odds of endophthalmitis, similar to that reported by Kiss et al ⁹ but conflicting with Bavinger and colleagues. ⁶ This association may reflect that clusters of sterile endophthalmitis secondary to aflibercept were reported during the study period ¹⁴ and may have inflated the occurrence of aflibercept-related endophthalmitis diagnoses. Additionally, the study period for aflibercept predated prefilled syringes and the safety advantage of such. ¹⁵

Owing to the nature of this medical claims data set, the study outcome was limited to the occurrence of endophthalmitis. Information related to disease course and visual acuity was not available. Similarly, we were not able to verify the presence of endophthalmitis with medical-record data or cultures. We acknowledge that prompt vitreous tap and injection of antibiotics and/or vitrectomy are paramount in the clinical care of endophthalmitis, and retina specialists are likely best positioned to administer prompt treatment given their training; however, we did not evaluate delay to treatment or how endophthalmitis was cared for among the ABO vs non-ABO cohorts or retina vs nonretina cohorts. Also, we did not evaluate other aspects of quality of care besides the risk of postinjection endophthalmitis among the cohorts.

Because ABO certification was collected at a single point in time, it is possible that some injections by early-career ophthalmologists were included as ABO certified even though their certification came later, or conversely, physicians who were ABO certified at the time of injection were not included because they had since lost certification. Similarly, retina subspecialty status was determined after the study period, and therefore some injections may have been administered prior to completion of a retina fellowship.

Additionally, we did not consider fellowship training in the retina field prior to residency, as sometimes occurs (eg, for foreign medical graduates). Also, our algorithm would not detect retina fellowship training for a physician who was unlisted in the ABO, AAO, ASRS, and Doximity directories and did not share training information on a clinic or hospital website, although we believe unlisted physicians to be rare. Also, we did not discern between 1-year medical retina fellowship training and 2-year medical/surgical retina fellowship training. Nevertheless, dilution between cohorts with regard to either ABO certification or retina fellowship status would have caused a bias to the null, and if this did exist, it would suggest our results underestimated the true safety effect of certification and training.

Despite nearly 3 million injections in our investigation, there were relatively few injections in the non–ABO-certified and nonretina cohorts, reflecting the US intravitreal injection landscape where the majority of providers in the United States are board certified and have completed a fellowship. ¹⁰ In turn, the low numbers in the non-ABO and the nonretina cohorts (64 and 61 endophthalmitis outcomes, respectively) limited statistical power for comparison between those in and out of the studied group. Indeed, our study was not sufficiently powered to distinguish a difference between the retina- and the nonretina-trained groups, possibly because of the small nonretina cohort, the rarity of endophthalmitis complications, and/or dilution between cohorts as discussed earlier. Nevertheless, our study demonstrated a significant finding for the ABO comparison. Meanwhile, the scope of practice for intravitreal injection is expanding in some states; it is possible that the non-ABO and nonretina cohorts will expand in the coming years, enabling a higher-powered study in the future. Presently, however, this study suggests that such a trend could have a negative impact on patient safety.

This study did not include the non-Medicare population and therefore had a lower proportion of younger patients, steroid injections, and non–age-related macular degeneration diagnoses (which tend to be higher risk for postinjection endophthalmitis) as compared with the general US population. We excluded patients without continuous Medicare coverage before and after intravitreal injection, patients with prior endophthalmitis diagnosis, and patients with ocular surgery within 14 days prior to injection to reduce inadvertently counting endophthalmitis events that were separate and unrelated to intravitreal injection. We further excluded injections with multiple and/or unknown provider numbers to minimize attributing injections to the wrong provider. We also excluded injections with missing J codes (designating medication) because they may have been corticosteroids, which are known to have higher odds of postinjection endophthalmitis. ⁸ Altogether, 435 318 (13%) of injections were excluded to reduce uncertainties associated with these injections.

In summary, this retrospective study of data from medical claims of intravitreal injections given to Medicare patients shows reduced odds of endophthalmitis for ophthalmologists who are ABO board certified. We conclude that certification is likely beneficial, especially for invasive procedures such as intravitreal injection.