Setting for Patient-Level Self-Matching Analysis

Ontario’s population at the study midpoint (2009) was 12 934 500, including 1 746 158 aged 65 years or older. Cerebrovascular disease accounted for 18% of total hospital admissions, more than any other single disease. In Ontario, intravitreal medication treatments for AMD and macular edema became popular in 2006 (bevacizumab) and fully funded in 2008 (ranibizumab). Over the study period, the rate of initiating intravitreal treatment equaled about 4.60 patients annually per 1000 individuals aged 65 and older. A past history of thromboembolic events was not an established contraindication to initiating intravitreal treatment (potential bias from withholding treatment owing to thromboembolic event history is explored in secondary analyses). The study was approved by the Sunnybrook Research Ethics Board prospectively and followed privacy guidelines of the Institute for Clinical Evaluative Sciences, including a waiver for consent.

Medical Care

Throughout the study, patients aged 65 years or older in Ontario were covered by universal health insurance that provided free access to physician services, diagnostic procedures, emergency visits, and prescription medications for outpatient and inpatient care. Patients could be tracked forward and backward in time through validated population-based databases. The health care databases included demographic information, physician fees, medication prescriptions, emergency contacts, and discharge diagnoses following hospitalization linked through unique coded patient identifiers. The accuracy of these databases has been previously validated. The available databases did not contain information on the specific indication for anti-VEGF treatment, contralateral eye disease, overall visual acuity, functional status, blood pressure, or anticoagulation compliance.

Patients

We identified consecutive patients initiating intravitreal injections between April 1, 2006 and March 31, 2013 by tracking health care database codes used in past research. The treatment date for each patient was defined as the day of the first intravitreal injection (OHIP billing codes E147 or E149). The 7-year enrollment period was selected to provide all data available since initiation of intravitreal anti-VEGF treatment for AMD and macular edema, with the exception of a few early adopters in 2005. The enrollment interval also provided a minimum of 1 year follow-up for all patients. We excluded patients under 65 years of age and those living outside of Ontario because of a lack of available data. In secondary analyses, we excluded the small number of patients who might have received intravitreal injections of antibiotics, steroids, or gas (explained below).

Patient Characteristics

Individual patient data ascertained on the treatment date ( Table 1 ) included age, sex, socioeconomic quintile (based on postal code average income), home location (urban or rural), and year of study enrollment. The patient ocular history included information on intraocular interventions in the prior 6 weeks (corneal, glaucoma, cataract, retina, and retinal laser), number of intravitreal injections, and cataract surgery in the prior decade. The presence of hypertension, diabetes, and emphysema was determined using validated algorithms, with special attention to diabetes in accord with past research. Health care utilization data were ascertained for the full year prior to initiating intravitreal treatment for total hospitalizations, emergency department visits, outpatient encounters (ophthalmology and non-ophthalmology), prescription medications, and AMD or macular edema diagnosis.

Specific Anti–Vascular Endothelial Growth Factor Medications

A patient was defined as receiving ranibizumab if a prescription for ranibizumab was filled within 2 weeks of initiating intravitreal treatment. The remaining patients were defined to have received bevacizumab (compounded off-label and not tracked by prescriptions). A separate prespecified subgroup analysis was also conducted to explore artifacts from fallible definitions by invoking a more stringent 2-step definition intended to reduce misclassification of treatment for uveitis, retinal detachment, or endophthalmitis. For this subgroup analysis, first, patients who had fewer than 3 intravitreal injections were excluded under the assumption that treatment for AMD or macular edema would consist of at least 3 loading injections. Second, patients who had an acute intraocular intervention (cataract, retinal, cornea, glaucoma surgery, or retinal laser) from 6 weeks prior to 1 week after the initial treatment date were excluded under the assumption that these intravitreal injections might be performed for other indications.

Acute Thromboembolic Events

We evaluated emergency department visits for thromboembolic events to test for acute unintended systemic consequences before and after initiating intravitreal anti-VEGF treatment. Physician diagnostic codes in the emergency department were used to identify the 3 most frequent specific thromboembolic emergencies; namely, ischemic stroke (codes I60–I64), acute myocardial infarction (codes I21, I22), and venous thromboembolism (codes I26, I80). The accuracy of these diagnostic codes has been previously validated. We did not stratify results by day of the week, season, or year owing to small sample sizes in subgroups. Other patient characteristics were subjected to subgroup analysis to test robustness.

Self-Matching Approach

Our analytic design (exposure crossover analysis) defined each patient as his or her own control since randomized trials have been insufficiently powered to assess toxicities and are ethically problematic, given that anti-VEGF treatment has become the standard of care. The design identified patients by the intervention and then tracked each patient’s outcomes forward and backward in time. A strength of the self-matching design was that it avoided all potential biases associated with selecting an independent control group. Similar to case-crossover analysis, the self-matching design removed confounding due to genetics, personality, education, and other stable characteristics (measured or unmeasured). Similar to time series analysis, an extended observational interval before and after intervention attenuated temporal confounders.

Time Intervals

Each patient’s 5-year time series data were divided into consecutive 28-day segments grouped into 3 larger intervals ( Supplemental Figure 1 , available at AJO.com ): the Baseline interval (year -4, -3, and -2 before first injection), the Induction interval (year -1 before first injection), and the Subsequent interval (year +1 after first injection). The main strength of this approach was in generating consecutive 1-month intervals before and after initiation of treatment for each individual patient. The 1-year Induction interval was defined in advance and considered a time when a patient’s functioning might have potentially differed from baseline. Serious medical events might be attenuated during the Induction interval before beginning ophthalmologic treatment; hence, the outcome rate during this period may be artificially distorted by selection bias and was not used to assess baseline risk.

Selection Bias

We assumed that patients might be less likely to receive anti-VEGF injections immediately following a thromboembolic event. As noted above, the analytic approach excluded the interval 1 year immediately prior to the first anti-VEGF injection to reduce such selection bias followed by regression to the mean. Thromboembolic risk naturally increases with age, thereby necessitating an extended baseline interval and possible general estimated equations models to adjust for anticipated (eg, age) and unanticipated temporal trends (eg, calendar year). Unlike population-level analyses, the analysis was patient-level and individually matched, thereby avoiding ecologic biases. Together, these analytic strategies helped to avoid spurious findings related to selection bias and temporal confounding.

Secondary Analyses

Further analyses explored more potential artifacts due to selection bias, assessed robustness, and checked for survivor bias. The first set examined a separate control group in a companion analysis, where the same analytic criteria were applied to identify patients undergoing first eye cataract surgery instead of intravitreal anti-VEGF therapy. We hypothesized cataract surgery would also be prone to selection bias, especially given the medical evaluation and anesthetic required for the operation. The second set explored other life-threatening emergencies, including unstable angina (code I20), congestive heart failure (code I50), hip fractures (code S72), depression (codes F32, F33), falls (codes W0, W1), cholecystitis (codes K80–K83), and total emergency visits among the patients receiving anti-VEGF treatment. To investigate survivor bias the third set excluded patients who did not survive the entire Subsequent interval (n = 2107).

One more set of secondary analyses stratified results based on the patient’s history of thromboembolic events (accounting for those with a remote history, a recent history, and remote or recent history of thromboembolic events). This analysis required defining year -3 and year -2 before initiating anti-VEGF injection as the Baseline interval, and then excluding patients who had a thromboembolic event in year -1, year -4, and year -4 or year -1 ( Supplemental Figure 3 , available at AJO.com ). This analysis addressed many other forms of selection bias, including the concern that ophthalmologists might delay treatment temporarily after an acute thromboembolic event.

Statistical Analysis

Our primary analysis evaluated emergency department visits for thromboembolic events and compared the patient’s Baseline interval to Subsequent interval. Statistical testing was calculated based on the McNemar approach and further explored using parsimonious longitudinal Poisson generalized estimating equations (GEE). A linear slope variable was included in the model when statistically significant (ie, outcomes with a trending baseline due to increasing age or other time trends). The Induction interval was examined only for descriptive purposes. A prespecified subgroup analysis was conducted to determine adverse event rate differences by the patient’s specific anti-VEGF medication and diabetic status. Post hoc subgroup analyses based on patient characteristics were used to assess robustness. Analyses were performed at the Institute for Clinical Evaluative Sciences using SAS version 9.3 (SAS Inc, Cary, North Carolina, USA).

Patient Characteristics

We identified 57 919 total patients. The patients were widely distributed over the full range of socioeconomic quintiles ( Table 1 ). About two-fifths received ranibizumab and about three-fifths received bevacizumab. Patients who received ranibizumab rather than bevacizumab were older, more likely to be female, more likely to have emphysema, less likely to have diabetes, and less likely to have undergone recent intraocular intervention. The 2 groups had similar mean numbers of total outpatient encounters, emergency visits, hospital admissions, and unique medication prescriptions during the year before initiating treatment.

Thromboembolic Events

A total of 3592 thromboembolic events occurred during the study interval, including 1858 during the 3-year Baseline interval and 1077 during the 1-year Subsequent interval ( Figure 1 ). The annual thromboembolic event rate per 1000 patients was 10.7 in the Baseline interval and 18.6 in the Subsequent interval. The absolute risk increase equaled 7.9 events per 1000 patients and was mathematically equivalent to a number needed to harm of 127 annually. The GEE model yielded a rate ratio of 1.74 (95% confidence interval [CI] 1.58–1.92; P < .0001). The thromboembolic events occurred at a median time of 58 days from last injection.

Thromboembolic emergencies after anti–vascular endothelial growth factor (VEGF) treatment and after cataract surgery. Each patient followed for thromboembolic emergencies resulting in an emergency visit. X-axis is divided into segments of 28 days, with time 0 defined as the patient’s first anti-VEGF treatment (Top) and first cataract surgery (Bottom). Y-axis shows the rate of emergency department visits for thromboembolic emergencies per 1000 patients annually. Induction interval denoted in orange is excluded from analysis owing to potential selection bias. Comparing the Baseline interval to the Subsequent interval, there was a significant increase in thromboembolic events after ant-VEGF treatment (Top), which was not apparent after cataract surgery (Bottom).

Specific Events

Of the 3 specific thromboembolic emergencies, acute ischemic stroke had the highest relative increase ( Table 2 , Figure 2 ). The rate per 1000 patients annually increased from 4.2 during the Baseline interval to 9.1 during the Subsequent interval (GEE-modeled rate ratio [RR] 2.18; 95% CI 1.94–2.46; P < .0001). The rate per 1000 patients annually for myocardial infarction increased from 3.1 to 5.1 (GEE-modeled RR 1.65; 95% CI 1.43–1.91; P < .0001) and for venous thromboembolism increased from 3.5 to 4.5 (GEE-modeled RR 1.29; 95% CI 0.99–1.69; P = .059).

Stroke emergencies and total emergencies after anti–vascular endothelial growth factor (VEGF) treatment. Each patient followed for medical event resulting in an emergency visit. X-axis divided into segments of 28 days, with time 0 defined as the patient’s first anti-VEGF treatment. Y-axis shows the rate of emergency department visits per 1000 patients annually. Induction interval denoted in orange is excluded from analysis owing to potential selection bias. Stroke emergencies (Top) and total emergencies (Bottom) are displayed. Comparing the Baseline interval to the Subsequent interval, there was a significant increase in stroke emergencies but not total emergencies after anti-VEGF treatment.

Control Group Companion Analysis

The control group companion analysis consisted of patients who underwent first eye cataract surgery (n = 471 429). A total of 25 880 thromboembolic events occurred during the study interval, including 13 977 during the 3-year Baseline interval and 6966 during the 1-year Subsequent interval ( Figure 1 ). The Baseline annual thromboembolic event rate per 1000 patients was 9.9, near identical to the Baseline rate among the patients receiving anti-VEGF treatment ( Figure 1 ). The time trend–adjusted GEE modeled rate ratio after cataract surgery was 1.10 (95% CI 0.97–1.24; P = .1372).

Other Life-Threatening Emergencies

No increase in Subsequent adverse event rates was apparent for unstable angina, hip fractures, falls, depression, or total emergency visits after anti-VEGF treatment. A small increase was observed for congestive heart failure and cholecystitis. The trending baseline was accounted for when a significant slope was identified ( P < .05); no significant slope was apparent for ischemic stroke, myocardial infarction, or venous thromboembolism ( Figure 2 , Supplemental Figure 2 , available at AJO.com ).

Selection Bias and Other Secondary Analyses

The higher rate of thromboembolic events during the Subsequent interval after anti-VEGF treatment was observed after stratifying by the patients’ demographic characteristics, ocular history, medical comorbidities, and specific intravitreal injection medication ( Table 3 ). The increased risk was similar in magnitude for patients enrolled during the early years and during the later years of the study. The increased risk persisted in the subgroup with no medical emergencies for any reason during the entire year before initiating treatment. The relative increase was accentuated among patients not diagnosed with hypertension or diabetes, as well as among those with fewer physician outpatient visits or those with fewer prescribed medications (perhaps suggesting some patients were not medically optimized).

Table 3

Thromboembolic Events During Baseline and Subsequent Interval in Patients Receiving Anti–Vascular Endothelial Growth Factor Treatment Stratified by Patient Characteristics

	Total Events	Baseline Rate a	Subsequent Rate a	Rate Ratio	95% Confidence Interval
Patient Demographics
Full cohort	3592	10.7	18.6	1.74	(1.58–1.92)
Age
65–74	954	9.0	15.4	1.71	(1.40–2.09)
75–84	1599	10.7	18.5	1.73	(1.51–1.99)
≥85	1039	13.2	23.5	1.78	(1.48–2.15)
Sex
Female	1945	10.1	18.1	1.80	(1.56–2.07)
Male	1647	11.5	19.2	1.68	(1.47–1.92)
Home status b
Urban	3062	10.5	18.3	1.74	(1.57–1.94)
Rural	530	12.0	20.7	1.72	(1.34–2.22)
Socioeconomic status c
Lowest	803	12.0	20.6	1.72	(1.42–2.08)
Next to lowest	775	10.6	18.2	1.72	(1.38–2.14)
Middle	726	11.4	18.6	1.63	(1.26–2.09)
Next to highest	663	9.9	19.5	1.97	(1.60–2.42)
Highest	612	9.6	15.8	1.65	(1.33–2.06)
Year of enrollment d
Early	803	9.9	19.2	1.94	(1.56–2.41)
Late	2789	10.9	18.4	1.69	(1.51–1.88)
Ocular History
Treatment regimen e , f
Short	1336	11.8	23.5	2.00	(1.69–2.36)
Sustained	2256	10.2	16.2	1.60	(1.42–1.80)
Recent intervention g
Yes	627	11.2	18.6	1.66	(1.34–2.07)
No	2965	10.6	18.6	1.76	(1.58–1.96)
Cataract surgery in past decade
Yes	2086	12.0	20.2	1.68	(1.47–1.91)
No	1506	9.2	16.8	1.84	(1.59–2.12)
Medical History
Hypertension e , h
Yes	3129	12.0	19.6	1.64	(1.48–1.82)
No	463	5.8	14.7	2.56	(1.93–3.40)
Diabetes e , h
Yes	1692	13.5	20.9	1.56	(1.36–1.78)
No	1900	8.9	17.1	1.92	(1.67–2.20)
Emphysema e , h
Yes	1135	13.8	22.4	1.63	(1.37–1.93)
No	2457	9.6	17.3	1.80	(1.60–2.02)
No. of outpatient visits e
≤6	157	6.0	14.8	2.40	(1.35–4.28)
≥7	3435	11.1	18.9	1.71	(1.55–1.88)
No. of ophthalmologist visits e
≤6	2703	10.5	18.7	1.77	(1.58–1.99)
≥7	889	11.2	18.4	1.65	(1.38–1.98)
Any emergency visit e
Yes	2145	14.9	24.3	1.63	(1.41–1.88)
No	1447	8.2	15.2	1.86	(1.63–2.13)
Any hospital admission e
Yes	1215	17.9	30.6	1.72	(1.38–2.14)
No	2377	9.4	16.4	1.75	(1.57–1.95)
Total drugs dispensed e
≤6	473	4.9	13.5	2.77	(2.10–3.66)
≥7	3119	12.8	20.4	1.60	(1.44–1.78)
Medication
Ranibizumab	1456	10.6	17.2	1.61	(1.39–1.87)
Bevacizumab i	2136	10.7	19.6	1.83	(1.61–2.09)

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

1. Subjectivity in Retinopathy of Prematurity Screening
2. Comparison of Clinical Outcomes of Iris Fixation and Scleral Fixation as Treatment for Intraocular Lens Dislocation
3. Association of Ocular Conditions With Narrow Angles in Different Ethnicities
4. Reporting Visual Acuities

Stay updated, free articles. Join our Telegram channel

Join