Yearly Treatment Patterns for Patients with Recently Diagnosed Diabetic Macular Edema Thiago A. Moulin, MD,1 Eric Adjei Boakye, PhD,2,3 Lorinette S. Wirth, MPH,1 Jiajing Chen, PhD,1 Thomas E. Burroughs, PhD,1 David E. Vollman, MD, MBA4 Purpose: To describe the treatment patterns and the predictors of different treatment standards in recently diagnosed diabetic macular edema (DME) patients in a nationally representative sample. Design: A retrospective cohort study using administrative claims data from January 1, 2007, through March 31, 2015. Patients were grouped into yearly cohorts. Participants: A total of 96 316 patients were included. Methods: Patients with a diagnosis of DME were identified using International Classification of Diseases, Ninth Edition, Clinical Modification, codes. Predictors of antievascular endothelial growth factor (VEGF) use and number of anti-VEGF injections per patient were assessed using generalized linear regression (logistic and negative binomial, respectively), and yearly trends in different treatments were analyzed with Mann-Kendall tests. Main Outcome Measures: Predictors of anti-VEGF treatment and of anti-VEGF injections per patient and the changes in relative use of DME therapies per cohort. Results: Among those with any treatment, the odds of being prescribed anti-VEGF therapy increased by 700% from 2009 to 2014 and by 154% for those seen by a retina specialist. Those in the cohort of year 2014 received 3.5 times more injections than those in 2009, whereas those covered by Managed Medicare, Medicaid, and Medicare received 31%, 24%, and 11% less injections. Anti-VEGF were 11.6% of all DME treatments in 2009 increasing to 61.9% in 2014, while corticosteroids and focal laser procedures dropped from 6.1% to 3% and 75% to 24%, respectively. Procedures per patient (PPP) were much lower than those observed in clinical trials of antiVEGF. Procedures per patient increased in the cases of aflibercept (from 1 in 2011 to 2.20 in 2014), bevacizumab (from 1.84 in 2009 to 3.40 in 2014), and ranibizumab (from 3.11 in 2009 to 4.48 in 2014), whereas applications of laser procedures and corticosteroids per patient remained roughly stable. Conclusions: Year of diagnosis and being seen by a retina specialist were important predictors of receiving anti-VEGF therapy, and after one received such therapy, the number of additional injections was smaller for those with government-provided insurance. Anti-VEGF therapy has become a mainstay in DME treatment, with PPP, although relatively low, also increasing. Ophthalmology Retina 2019;3:362-370 ª 2018 by the American Academy of Ophthalmology
The World Health Organization calculates that 422 million people live with diabetes mellitus (DM) worldwide.1 In the United States, the Centers for Disease Control and Prevention estimates that more than 30 million Americans are diabetic.2 Diabetic macular edema (DME) is one of the developments in DM that can threaten vision and, in 2010 alone, affected 746 000 Americans,3 increasing healthcare costs4 and lowering the quality of life of those individuals. Treatment of DME consists of focal laser photocoagulation (FLP), corticosteroids, and, since the mid 2000s, antievascular endothelial growth factor (VEGF) therapy, which has become first-line therapy for the disease5 given its superior safety and efficacy. Innovation also has produced newer, better corticosteroid implants approved by the Food and Drug Administration in 2014.
362
2018 by the American Academy of Ophthalmology Published by Elsevier Inc.
These important changes in DME treatment have been studied recently by different authors6e14; however, only 3 studies used large national databases, 2 studies11,12 looking at data up to 2011 and a third study13 looking at data as recent as 2015, examined exclusively anti-VEGF therapy and did not include Medicaid or non-Advantage Medicare patients. It was our goal to complement and update this knowledge by analyzing a national healthcare database with public and private insurance claims from January 2007 through May 2015, describing and comparing anti-VEGF, corticosteroid, and FLP for recently diagnosed DME. The first aim was to describe the predictors of receiving antiVEGF treatment; the second was to present the predictors of the number of injections, given that patients received antiVEGF therapy at least once; and a third was to analyze the evolution of treatment by year of diagnosis by looking at the
https://doi.org/10.1016/j.oret.2018.11.014 ISSN 2468-6530/19
Moulin et al
Treatment Patterns for Recently Diagnosed DME Table 1. Codes Used in This Study
shares of treatments and the number of procedures per patient (PPP).
Codes
Methods Data Source, Study Design, and Study Sample The present study used de-identified health administrative claims data provided by Symphony Health Solutions (Horsham, PA) on more than 8.7 million unique diabetic patients with commercial or government-provided health insurance from January 1, 2007, through March 31, 2015. The data contain information on patient age and gender, diagnoses (as coded in the International Classification of Diseases, Ninth Edition, Clinical Modification [ICD-9CM]), inpatient and outpatient procedures (coded in Current Procedural Terminology [CPT] and the Healthcare Common Procedure Coding System [HCPCS]), insurance plans, and providers (Table 1). Saint Louis University Institutional Review Board approved this research and patient consent requirement was waived given we studied de-identified patient information. Using a retrospective cohort design, data were divided into yearly cohorts and patients were followed up for 1 year after their index date, defined as the date of their first insurance claim with a diagnosis of DME, ensuring that we indexed patients only once. For example, patients who were indexed on November 20, 2010, were included in the 2010 cohort, but were followed up until November 19, 2011. Inclusion criterion was a diagnosis of DME, defined as a visit with either an ICD-9-CM code of 362.07 (“diabetic macular edema”) or a visit with a composite diagnosis of DM and of macular edema. For the latter, we used a previously validated algorithm15 with one modification: to satisfy the condition of a diagnosis of DM, we included not only those with 250.xx ICD9-CM codes, as in the original algorithm, but also those with a DR diagnosis (ICD-9-CM codes 362.01e362.06). We believe this may capture DME patients seen by ophthalmologists more comprehensively, given that some of these professionals may be more familiar with the codes for ocular diabetes versus systemic DM. To ensure proper continuity and integrality of treatment and that cases are truly newly diagnosed DME,16 patients were excluded if they had been enrolled in the plan for less than 2 years before or less than 1 year after the index date. Several strategies were used to ensure that the resources being studied indeed were used for DME treatment: (1) patients who had a diagnosis, in any point in time, of diseases that may lead to confusion of DME diagnosis were excluded, a procedure performed in a prior work12; (2) patients with a diagnosis of neoplasias who also were treated with bevacizumab were excluded, as also was done previously11; (3) procedures of interest were captured only if they were billed with DME as the associated diagnosis or if the procedure matched a visit billed with DME as diagnosis that happened on the same day, with the same physician. Assessment of procedure-associated DME diagnosis was carried out as explained above, through a singular and a composite coding algorithm. Drugs of interest were aflibercept (Eylea; Regeneron Pharmaceuticals, Inc, Tarrytown, NY), bevacizumab (Avastin; Genentech, South San Francisco, CA), dexamethasone intravitreal implant (Ozurdex; Allergan, Irvine, CA), 0.19-mg fluocinolone acetonide intravitreal implant (Iluvien; pSivida Corp, Watertown, MA), pegaptanib (Macugen; Eyetech Pharmaceuticals, New York, NY), ranibizumab (Lucentis; Genentech), and triamcinolone acetonide injections. Unspecified drugs were injections (Current Procedural Terminology code 67028) not paired with a corresponding HCPCS drug code. Given that the years being observed in our study saw
ICD-9-CM codes 362.07, 362.53 þ (250.xx / 362.01 / 362.02 / 362.03 / 362.04 / 362.05 / 362.06) 362.15, 362.16, 362.2x, 364.42, 365.63, 365.89 282.6 362.3x 362.4x 362.5x 363.xx, 364.0x, 364.1x, 364.2x, 364.3x, 364.4x, 365.1x, 365.2x, 365.3x, 365.4x, 365.5x, 365.6x, 365.7x 153.9 162.9 180.9 191.9 CPT codes 67028 67210
HCPCS codes J2503 C9257 J9035 J7999* J2778 C9233 J0178 C9291 Q2046 J3300 J3301 C9450 J7313 J7311 J7312 J3490 J3590 C9399
Description Diabetic macular edema Proliferative retinopathies Sickle cell disease Vein occlusions Separation of retinal layers Macular degeneration Uveitis Glaucoma Malignant neoplasm Malignant neoplasm and lung Malignant neoplasm uteri Malignant neoplasm
of colon of bronchus of cervix of brain
Intravitreal injection of a pharmacologic agent Destruction of localized lesion of retina (e.g., macular edema, tumors), 1 or more sessions; photocoagulation Pegaptanib Bevacizumab Bevacizumab Bevacizumab* Ranibizumab Ranibizumab Aflibercept Aflibercept Aflibercept Triamcinolone injection Triamcinolone injection 0.19-mg fluocinolone implant 0.19-mg fluocinolone implant 0.59-mg fluocinolone implant Dexamethasone implant Unspecified drug Unspecified drug Unspecified drug
CPT ¼ Current Procedural Terminology; HCPCS ¼ Healthcare Common Procedure Coding System; ICD-9-CM ¼ International Classification of Diseases, Ninth Edition, Clinical Modification. *CPT code J7999 actually refers to “Compounded drug, not otherwise classified,” but is often used to code for bevacizumab intravitreal injections.
the introduction of several new therapies and the fact that permanent HCPCS codes are not issued immediately on drug commercialization, drugs such as aflibercept or the corticosteroid implants probably were captured partly by this category. Aflibercept, for example, was granted its permanent HCPCS code only in 2013, having 2 different codes during 2012 and having no specific code before April 2012. Another possibility is that bevacizumab, a compounded drug without a specific HCPCS code for eye use, was coded in ways we have not predicted. Studying these codes was
363
Ophthalmology Retina Volume 3, Number 4, April 2019 Medicare Advantage plans; Medicaid; Medicare; other; third party, which encompasses private insurance plans; and unknown. A patient’s insurance plan type was defined as the plan type of the claim with the closest date to such patient’s index date. The category “other” originally included 2006 claims to which we added 959 entries that initially were pairs of claims for the same event, but with 2 different plan types, and thus were collapsed into 1 claim with an “other” plan type. Most of these 959 entries initially were duplicates, where one was Medicare and the other third party, for patients who had supplemental Medicare coverage. Treatment standard was defined as anti-VEGF if a patient received any antiVEGF treatment during the follow-up; as other treatment only if they received any treatment, but not anti-VEGF; or as not treated.
Statistical Analysis
Figure 1. Patient selection flow chart. *Related diagnoses were those that may lead to diagnostic uncertainty regarding diabetic macular edema (DME) or that require similar treatments.
important in determining a treatment standard, although they were not considered anti-VEGF treatment, but rather other treatment. Table 1 contains all the diagnosis, procedure, and drug codes used in our study.
Measures The main outcome variable was whether a patient received any anti-VEGF treatment during first year of diagnosis given that any treatment was dispensed. It is important to study anti-VEGF prescriptions in the treated subset of the population because diagnosis of DME does not necessarily indicate treatment, for example in cases where there is no significant visual impairment or the edema is far away enough from the fovea. Secondary outcomes were number of additional anti-VEGF injections, given that at least 1 anti-VEGF treatment was received and the variation of treatment methods’ share and PPP, which was calculated as the mean of the number of times a patient received a given treatment during our 1-year follow-up. Independent variables included age, gender, number of DME office visits (for the main outcome only), whether the provider was a retina specialist, insurance plan type, and year of diagnosis. Office visits were calculated by counting only visits billed with a DME diagnosis. The purpose of the visit was of no interest in this calculation; thus visits for injections, laser procedures, OCT scans, and so forth all were included in this count. Retina specialist status was incorporated to providers’ information by matching first and last name and the first 3 digits of the zip code from our dataset to a list obtained from the American Society of Retina Specialists’ website.17 Insurance plan types were categorized as: cash, for those claims paid with out-of-pocket funds; managed Medicare, for claims billed to patients with
364
For the first aim, we used a logistic regression model to assess the factors associated with receiving any anti-VEGF during the first year of diagnosis among those who received at least 1 of the treatments we studied (anti-VEGF, corticosteroids, and FLP). For the second aim, we used a negative binomial regression model on the counts of additional anti-VEGF injections among those who had at least 1 prescribed. Therefore, those who received just 1 injection had an incidence rate (the number of additional injections in 1 year) of 0. A quantileequantile plot was used to assess fitness of a binomial distribution to our data. Model variables were selected based on model quality as shown by the Akaike information criterion. Model fitness was evaluated with residual versus fitted plots. Descriptive statistics and trend analysis with MannKendall tests were used to compare yearly cohorts, as set out in the third aim. Trends were tested on means for continuous variables and on frequencies for categorical variables. SAS software version 9.0 (SAS Institute, Inc, Cary, NC) was used for data extraction, and R software version 3.5 (R Development Core Team, Vienna, Austria) was used for data analysis and graphing. Statistical significance was set at a ¼ 0.05. Missing values were handled through list-wise deletion given the large sample size and suspected randomness of lack of data points.
Results A total of 401 952 patients with a diagnosis of DME were found in our dataset. This study encompassed a final sample of 96 316 patients after exclusion criteria were applied. Figure 1 depicts our patient selection process. Extraction of data from the American Society of Retina Specialists’ website resulted in a list of 1711 retina specialists residing in the United States. Of these, 880 were matched to providers of our study sample, a match rate of 51% (in total, 1332 of the 1711 specialists provided to patients in the larger, 8.7-millionpatient dataset). Because patients were required to be enrolled in the same plan 2 years before and 1 year after indexing, there were no cohorts in years 2007, 2008, and 2015, which resulted in a total of 6 cohorts being formed. Also, for the same reason, the cohort of year 2014 is reduced in number given that we had data through March 31, 2015; therefore, patients enrolled after March 31, 2014, were not included. There was a statistically significant monotonic trend throughout cohort years in some cohort characteristics, namely gender (P ¼ 0.02), number of office visits (P < 0.01), treatment standards, and insurance plan types managed Medicare (P < 0.01), Medicaid (P ¼ 0.02), Medicare (P ¼ 0.02), and other (P < 0.01), with an increase in all of these but Medicare. Plan
Moulin et al
Treatment Patterns for Recently Diagnosed DME
Table 2. Characteristics of Our Study Sample Cohort Year Characteristic
2009
2010
2011
2012
2013
2014
P Value*
No. of participants Age (yrs), mean (SD) Female gender, no. (%) Office visits, mean (SD) Insurance plan type, no. (%) Third party Cash Managed Medicare Medicaid Medicare Other Unknown Retina specialist yes, no. (%) Treatment standard, no. (%) Anti-VEGF treatment Not treated Other treatment only
11 783 60.23 (10.77) 6069 (51.5) 2.20 (1.75)
16 577 59.89 (11.10) 8370 (50.5) 2.44 (2.13)
18 292 59.65 (11.40) 8991 (49.2) 2.56 (2.30)
20 329 59.70 (11.42) 10 125 (49.8) 2.92 (2.82)
23 833 59.67 (11.48) 11 660 (48.9) 3.18 (3.11)
5502 59.59 (11.53) 2685 (48.8) 3.60 (3.52)
0.027 0.060 0.024 0.009
7941 14 52 556 3106 99 15 3182
11 353 15 78 731 4230 148 22 4308
12 628 31 123 910 4356 200 44 4888
14 025 23 173 1318 4335 346 109 5309
16 259 10 262 1611 5200 475 16 6227
3801 2 79 446 1038 136 0 1420
(69.1) (0.0) (1.4) (8.1) (18.9) (2.5) (0.0) (25.8)
0.260 0.133 0.009 0.024 0.024 0.009 0.707 0.260
(67.4) (0.1) (0.4) (4.7) (26.4) (0.8) (0.1) (27.0)
(68.5) (0.1) (0.5) (4.4) (25.5) (0.9) (0.1) (26.0)
(69.0) (0.2) (0.7) (5.0) (23.8) (1.1) (0.2) (26.7)
(69.0) (0.1) (0.9) (6.5) (21.3) (1.7) (0.5) (26.1)
(68.2) (0.0) (1.1) (6.8) (21.8) (2.0) (0.1) (26.1)
593 (5.0)
1520 (9.2)
2481 (13.6)
3702 (18.2)
5355 (22.5)
1490 (27.1)
0.009
6577 (55.8) 4613 (39.1)
8927 (53.9) 6130 (37.0)
9583 (52.4) 6228 (34.0)
10 591 (52.1) 6036 (29.7)
12 452 (52.2) 6026 (25.3)
2755 (50.1) 1257 (22.8)
0.024 0.009
SD ¼ standard deviation; VEGF ¼ vascular endothelial growth factor. *Mann-Kendall monotonic trend tests, performed on means for continuous variables and on percent shares for categorical variables.
types cash, third party, and unknown; age; and retina specialists status did not exhibit a statistically significant monotonic trend (P > 0.05). Cohorts’ numbers steadily increased from 11 783 patients in 2009 to 23 833 patients in 2013. The mean number of DME-related office visits increased from 2.20 (standard deviation [SD], 1.75) to 3.60 (SD, 3.52). The relative number of patients receiving no treatment or treatment that is not anti-VEGF also decreased along the 6 years observed, going from 55.8% to 50.1% (P ¼ 0.02) and 39.1% to 22.8% (P < 0.01), respectively. The proportion of patients receiving any anti-VEGF during the 1-year observation increased from 5% in 2009 to 27.1% in 2014 (P < 0.01; Table 2). Regarding the treatments, anti-VEGF agents accounted for 11.6% of therapies in 2009, going up to 61.9% in 2014, whereas corticosteroids and FLP procedures decreased their participation, dropping from 6.1% to 2.8% for the former and 75.3% to 24.0% for the latter. A statistically significant monotonic trend was observed across all treatment types with the exception of unspecified drug. Although prescriptions of aflibercept (P ¼ 0.01), bevacizumab (P < 0.01), ranibizumab (P < 0.01), and dexamethasone implants (P ¼ 0.01) showed an increasing trend, pegaptanib (P ¼ 0.02), FLP (P < 0.01), and triamcinolone injections (P < 0.01) showed a decreasing trend. In our data, 0.19-mg fluocinolone implant-specific codes were not observed. Although FLP share of treatments decreased incisively during the study period, its absolute numbers increased from 7643 in 2009 to 11 489 in 2013. The number of unspecified drugs billed as intravitreal injections increased from 719 in 2009 to 5826 in 2013, with relative frequency peaking in 2012 at 17.4%, up from 7.1% in 2009, and then decreasing to 11.3% in 2014. The proportional participation of each treatment can be seen in Figure 2.
Figure 3 depicts the number of mean PPP by cohort year segmented by different plan types. Insurance plan types cash, other, and unknown were collapsed into “other plan types”. Analyzing trends among all patients, regardless of plan types, aflibercept (P ¼ 0.02), bevacizumab (P < 0.01), and unspecified drug (P < 0.01) PPP followed a monotonic trend, all positive. Ranibizumab went from 3.26 (SD, 2.58) to 4.74 (SD, 3.84), demonstrating a decrease in 2011, when its PPP was only 2.50 (SD, 2.26), which is why this trend testing was not significant. Ignoring the year 2011, ranibizumab does indeed exhibit a positive monotonic trend (P ¼ 0.02). Aflibercept went from 1 PPP in 2011 (only 1 injection in 1 patient) to 2.37 (SD, 1.78) in 2014, bevacizumab increased from 1.88 (SD, 1.47) to 3.66 (SD, 3.22) PPP from 2009 to 2014, whereas dexamethasone implants, triamcinolone injections, and FLP showed little variation throughout the years, ranging from 1 to 1.73, 1.42 to 1.65, and 1.63 to 1.68, respectively. Logistic regression was performed to study the predictors of receiving any anti-VEGF treatment, among those who were prescribed any treatments, with results displayed in Table 3. The odds of being prescribed an anti-VEGF treatment in the first year of diagnosis of DME increased substantially year after year, with those diagnosed in 2014 having an 8-fold (confidence interval [CI], 6.02e9.47) increase in the odds of being prescribed anti-VEGF therapy relative to those diagnosed in 2009. Female gender had a slightly negative effect, accounting for 6% decreased odds (CI, 0.90e0.98). Number of office visits, used in this model as a surrogate for patient adherence to medical care, increased these odds in 30% (CI, 1.29e1.32), whereas being seen by a retina specialist at least once during the first year of DME diagnosis increased in 154% (CI, 2.12e3.04) the odds of being prescribed anti-VEGF therapy. We also included an interaction term for retina specialist
365
Ophthalmology Retina Volume 3, Number 4, April 2019
Figure 2. Graph showing relative frequency of treatments by cohort. *Mann Kendall (MK) test 0.1 > P > 0.05. **MK test 0.05 > P > 0.01. ***MK test P < 0.01. VEGF ¼ vascular endothelial growth factor.
status and cohort year in our models, suspecting that over time the popularization of anti-VEGF treatment among all practitioners would reduce the influence of retina specialist status on prescriptions of this medication class. Table 3 shows that the effect of being a retina specialist (odds ratio [OR], 2.54) in the 2009 cohort was diminished along the cohort years, being multiplied by the interaction term. For example, for those in the 2013 cohort, the effect of being seen by a retina specialist resulted in an OR of only 1.30 (2.54 0.51). Overall, there is a trend toward decreased influence of one’s physician being a retina specialist on prescription of anti-VEGF during the first year of DME diagnosis. A second logistic regression analysis was performed to study the predictors of the number of additional anti-VEGF injections received after the first one, during the first year of diagnosis of DME, also summarized in Table 3. Those in the 2014 cohort showed an incidence rate ratio of 3.5 (CI, 2.90e4.30), meaning that, holding the other variables constant, those in the 2014 cohort, compared with those in the 2009 cohort, are expected to have a 3.5 times greater number of additional anti-VEGF injections. Those with managed Medicare, Medicaid, and Medicare plans showed reduced incidence rate ratios when compared with third-party insurance holders by 31% (CI, 0.55e0.87), 24% (CI, 0.70e0.83), and 10% (CI, 0.85e0.95), respectively. The
366
number of office visits was not used as a covariate in this analysis because of overfitting, as evidenced by a P value of 0 and an increase in the Akaike information criterion. This should not be surprising given that the counts of office visits were calculated by taking into account the visits resulting from the treatments that we are now using as the outcome in our model.
Discussion This retrospective claims database analysis studied a nationally representative sample consisting of 96 316 patients recently diagnosed with DME and their treatments during the first year of diagnosis. The number of patients diagnosed increased over the years, in a trend consistent with a similar study.12 Modifications in therapy for the disease are quite clear in the data, demonstrating an overall upward trend for anti-VEGF use, whereas FLP decreased its relative participation and corticosteroids maintained a consistent, albeit small, share. The share of patients left untreated declined, from 55.8% to 50.1%. Among those who were prescribed a treatment, being diagnosed in more recent years, going to the doctor’s office more often, and being seen at least once by a retina specialist increased one’s odds
Moulin et al
Treatment Patterns for Recently Diagnosed DME
Figure 3. Graphs showing procedures per patient (mean number of times a patient received a given treatment) by different plan types. *Mann Kendall (MK) test 0.1 > P > 0.05. **MK test 0.05 > P > 0.01. ***MK test P < 0.01.
of receiving anti-VEGF. After being prescribed such therapy, the incidence rate of receiving additional injections was greater for those diagnosed in later years and those who possessed private insurance. There are several factors that may help us to understand the increase in DME diagnosis over the years: first, more diagnoses may be the result of increased surveillance of the disease, a finding in accord with a large study on diabetic retinopathy screening rates between 2001 and 2013.18 It also could be a result of the concurrent popularization of OCT methods during the 2000s,19 the increased sensitivity of OCT, or both. Finally, more diagnoses may be the result of the availability of better treatments. Patients may be more willing to seek (and primary care practitioners to refer) ophthalmologic care knowing that the risk-to-benefit ratio of current treatment is higher. Diabetic macular edema does not necessarily prompt treatment; therefore, we analyzed the predictors of receiving anti-VEGF treatment only among those treated. A stepwise increase in roughly 2 points in OR per year was noted, as more evidence mounted supporting the use of anti-VEGF agents from several large studies.20e24 Another important predictor was being seen by a retina specialist at least once during the first year of DME diagnosis, which increased the odds of anti-VEGF prescription by 2.5 times, suggesting that these professionals switched to these therapies as firstline treatment faster than their counterparts. However, this effect declined over time, with those in the 2014 cohort who saw a retina specialist having an increased OR of just 1.42
times. It is possible that initially nonretina specialist ophthalmologists were not comfortable with injecting these drugs and thus referred patients to retina specialists, inflating those odds. It is also possible that most ophthalmologists would not prescribe a therapy that was still being studied and that was not yet fully endorsed by the American Academy of Ophthalmology’s Preferred Practice Pattern, which did so in 2016.5 We were also interested in studying the correlates of the number of anti-VEGF injections received. For this, we studied the incidence rate of additional injections during the first year of DME diagnosis among those who received at least 1 injection of this medication class. The number of injections increased yearly, with those in 2014 having 3.5 times more injections than those in 2009, which also can be seen in the number of injections per patient exhibited in Figure 3. Similarly to the discussion above, this increment can be explained by an increased understanding, by patients and practitioners, of anti-VEGF in DME. It is possible that in the first years, ophthalmologists would associate FLP earlier or more often with anti-VEGF therapy (which would reduce the number of injections dispensed), changing this practice when publications such as the protocol I and its follow-up studies20,25,26 demonstrated the superiority of ranibizumab plus prompt or deferred laser, going as far as suggesting that, for some patients, ranibizumab alone is the best option. This analysis of counts of anti-VEGF injections also provided evidence that those covered entirely by private insurance (third-party insurance)
367
Ophthalmology Retina Volume 3, Number 4, April 2019 Table 3. Predictors of AntieVascular Endothelial Growth Factor Prescription Outcome Received Any AntieVascular Endothelial Growth Factor
Counts of Additional AntieVascular Endothelial Growth Factor Injections
Confidence Interval Predictor Cohort (reference ¼ 2009) 2010 2011 2012 2013 2014 Retina specialist? (reference ¼ no) Insurance plan type (reference ¼ third party) Cash Managed Medicare Medicaid Medicare Other Unknown Gender (reference ¼ male) Age Interaction terms Cohort 2010: retina specialist Cohort 2011: retina specialist Cohort 2012: retina specialist Cohort 2013: retina specialist Cohort 2014: retina specialist Office visits
2.50%
97.50%
Incidence Rate Ratio
2.50%
97.50%
2.04 3.63 5.13 6.88 8.01 2.54
1.76 3.16 4.48 6.02 6.83 2.13
2.37 4.19 5.90 7.88 9.42 3.05
1.76 1.90 2.49 2.83 3.53 1.09
1.44 1.57 2.07 2.35 2.90 0.85
2.15 2.30 3.00 3.40 4.30 1.39
0.97 1.22 1.22 0.91 1.25 0.89 0.94 1.00
0.43 0.96 1.12 0.86 1.05 0.55 0.90 0.99
2.14 1.55 1.34 0.97 1.49 1.41 0.98 1.00
1.49 0.69 0.76 0.90 1.00 0.52 1.01 0.99
0.77 0.55 0.70 0.85 0.86 0.30 1.00 0.95
3.15 0.87 0.83 0.95 1.17 0.89 1.01 1.03
0.74 0.55 0.49 0.51 0.56 1.30
0.60 0.45 0.40 0.42 0.43 1.29
0.92 0.68 0.60 0.62 0.72 1.32
0.78 0.59 0.91 0.69 1.03 0.79 1.01 0.78 0.97 0.73 Not included because of overfitting
had 45%, 31%, and 12% more injections than those in managed Medicare, Medicaid, and Medicare plans, respectively, which may be the result of reduced medication coverage in these plans when compared with private insurance.27 More research should be carried out to investigate this possible gap and to study ways to decrease this disparity, given that the lower frequency of injections has been shown to be associated with worse outcomes.7,10 Figure 3 depicts the evolution of PPP over the years by different plan types, which should be of importance taking into account the discrepancy presented above and the fact that these years saw the passage of the Affordable Care Act (in 2010) and the Medicaid expansion in some states in the ensuing years. Overall, the number of anti-VEGF injections per patient from 2009 through 2014 was significantly less than what is predicated in the large randomized controlled trials that recommend their use. Whereas these studies20e24,28,29 administered from 7 to 12 injections per patient per year, real-world studies report much lower rates, ranging from 2 to 4 injections per patient per year,7,10e12 in line with the 1.0 to 4.7 PPP presented in this study (depending on the drug and cohort). The reason behind such a low PPP rate is not clear. As already stated, one reason may be practitioners’ adjustment to this new treatment regimen and the ongoing research in the field being published during the years of this study. Another possibility is a decreased compliance to therapy given the higher disease burden observed among DME versus DM patients.30,31 It is also possible that practitioners opted to switch therapies
368
Confidence Interval
Odds Ratio
1.03 1.18 1.33 1.30 1.27
after an inadequate anti-VEGF response, a decision supported by a panel of experts in the disease.32 Finally, another reason may be that partial, sufficient responses coupled with financial and plan limitations contributed to a reduced PPP when compared with clinical trials: the patient might have had a “good enough” response and, faced with another copayment, decided against an injection. Findings from the current study also suggest that ranibizumab injections are given more frequently than bevacizumab injections, especially in those with private insurance, which may be a result of the Food and Drug Administration approval and thus more widespread insurance reimbursement of the former, the complications inherent to using a compounded drug such as the latter, or both. It is also possible that those using bevacizumab experienced higher rates of therapy change. Overall, however, bevacizumab was the number 1 drug in terms of relative shares and in absolute number of claims, in accordance with prior works.7e9,11,12 Retrospective claims database analysis cannot evaluate chart-level clinical data. This makes associations between usage and real necessity of treatments impossible. There is also the possibility of coding errors, which could be the cause for 5012 intravitreal injections not being paired with any of the drugs studied and may be the cause of other unaccounted errors. Also, as in previous DME studies,7,12 our dataset did not provide treatment per eye; therefore, we cannot make conclusions regarding the treatment of an eye, but of a patient with at least 1 eye with DME. The prevalence of the outcomes from our logistic regression (16%) was more than 10%; therefore, the ORs presented
Moulin et al
Treatment Patterns for Recently Diagnosed DME
here are an overestimation of the true risk ratios, the appropriate association measure for cohort studies.33 Identification of a patient’s first diagnosis also is imperfect, because some patients might have been diagnosed in dates outside of our data or had their unique identifier changed during the study period, which can happen if they change plans, for example. Retina specialist status was an approximation, because we identified only 78% of all American Society of Retina Specialists members in our dataset and matched the information using first and last names and the first 3 digits of their zip codes, and not unique physician identifiers because these were not available. Furthermore, some retina specialists are not affiliated with the American Society of Retina Specialists. In any case, the effect of that variable in our analysis could be overestimated given that being seen by a specialist once during 1 year does not mean that one’s treatment was wholly conducted by a retina specialist. Furthermore, the effect in treatment standards of being seen by a specialist might be more of a function of disease severity than practitioner influence. Finally, it is also important to note that 19 219 injections were coded as unspecified drugs, with aflibercept, bevacizumab, and possibly corticosteroids being captured in that category. The results of our study were able to provide a clearer understanding of how DME treatment has evolved in the past decade, with anti-VEGF therapy becoming a mainstay. Although this is an improvement for patients, our study also suggested that DME is likely undertreated in the real world, with those with nonprivate insurance receiving even fewer injections. Future research investigating whether undertreatment is indeed a reality and its causes should be of great value. References 1. Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. Samet J, ed. PLoS Med. 2006;3:2011e2030 2. Centers for Disease Control and Prevention. National Diabetes Statistics Report, 2017. Estimates of diabetes and its burden in the United States background. https://www.cdc.gov/diabetes/ pdfs/data/statistics/national-diabetes-statistics-report.pdf; 2017. Accessed July 19, 2018. 3. Varma R, Bressler NM, Doan QV, et al. Prevalence of and risk factors for diabetic macular edema in the United States. JAMA Ophthalmol. 2014;132:1334e1340. 4. Shea AM, Curtis LH, Hammill BG, et al. Resource use and costs associated with diabetic macular edema in elderly persons. Arch Ophthalmol. 2008;126:1748e1754. 5. American Academy of Ophthalmology Retina/Vitreous Panel. Preferred Practice Pattern Guidelines. Diabetic Retinopathy. San Francisco, CA: American Academy of Ophthalmology; 2017. Available at: www.aao.org/ppp. Accessed July 19, 2018. 6. Feman SS, Chen J, Burroughs TE. Change in diabetic panretinal photocoagulation incidence. Ophthalmic Surg Lasers Imaging. 2012;43:270e274. 7. Kiss S, Liu Y, Brown J, et al. Clinical utilization of antivascular endothelial growth-factor agents and patient monitoring in retinal vein occlusion and diabetic macular edema. Clin Ophthalmol. 2014;8:1611e1621.
8. Blinder KJ, Dugel PU, Chen S, et al. Anti-VEGF treatment of diabetic macular edema in clinical practice: effectiveness and patterns of use (ECHO study report 1). Clin Ophthalmol. 2017;11:393e401. 9. Jusufbegovic D, Mugavin MO, Schaal S. Evolution of controlling diabetic retinopathy: changing trends in the management of diabetic macular edema at a single institution over the past decade. Retina. 2015;35:929e934. 10. Fong DS, Luong TQ, Contreras R, et al. Treatment patterns and 2-year vision outcomes with bevacizumab in diabetic macular edema: an analysis from a large U.S. integrated health care system. Retina. 2018;38:1830e1838. 11. Jiang S, Barner JC, Park C, Ling Y-L. Treatment patterns of anti-vascular endothelial growth factor and laser therapy among patients with diabetic macular edema. J Manag Care Spec Pharm. 2015;21:735e741. 12. VanderBeek BL, Shah N, Parikh PC, Ma L. Trends in the care of diabetic macular edema: analysis of a national cohort. Virgili G, ed. PLoS One. 2016;11:e0149450 13. Parikh R, Ross JS, Sangaralingham LR, et al. Trends of antivascular endothelial growth factor use in ophthalmology among privately insured and medicare advantage patients. Ophthalmology. 2017;124:352e358. 14. Wu CM, Wu AM, Greenberg PB, et al. Frequency of bevacizumab and ranibizumab injections for diabetic macular edema in medicare beneficiaries. Ophthalmic Surg Lasers Imaging Retin. 2018;49:241e244. 15. Bearelly S, Mruthyunjaya P, Tzeng JP, et al. Identification of patients with diabetic macular edema from claims data. Arch Ophthalmol. 2008;126:986e989. 16. Stein JD, Blachley TS, Musch DC. Identification of persons with incident ocular diseases using health care claims databases. Am J Ophthalmol. 2013;156:1169e1175.e3. 17. The American Society of Retina Specialists. Find a retina specialist. https://www.asrs.org/find-a-specialist/results/List? searchName¼&searchLocation¼. Accessed May 31, 2018. 18. Fitch K, Weisman T, Engel T, et al. Longitudinal commercial claims-based cost analysis of diabetic retinopathy screening patterns. Am Heal Drug Benefits. 2015;8: 300e308. 19. Fujimoto J, Swanson E. The development, commercialization, and impact of optical coherence tomography. Invest Ophthalmol Vis Sci. 2016;57:OCT1eOCT13. 20. Elman MJ, Aiello LP, Beck RW, et al. Randomized trial evaluating ranibizumab plus prompt or deferred laser or triamcinolone plus prompt laser for diabetic macular edema. Ophthalmology. 2010;117:1064e1077.e35. 21. Do DV, Schmidt-Erfurth U, Gonzalez VH, et al. The da VINCI study: phase 2 primary results of VEGF Trap-Eye in patients with diabetic macular edema. Ophthalmology. 2011;118:1819e1826. 22. Michaelides M, Kaines A, Hamilton RD, et al. A prospective randomized trial of intravitreal bevacizumab or laser therapy in the management of diabetic macular edema (BOLT study). 12-month data: report 2. Ophthalmology. 2010;117: 1078e1086.e2. 23. Nguyen QD, Brown DM, Marcus DM, et al. Ranibizumab for diabetic macular edema: results from 2 phase III randomized trials: RISE and RIDE. Ophthalmology. 2012;119:789e801. 24. Mitchell P, Bandello F, Schmidt-Erfurth U, et al. The RESTORE study: ranibizumab monotherapy or combined with laser versus laser monotherapy for diabetic macular edema. Ophthalmology. 2011;118:615e625. 25. Elman MJ, Bressler NM, Qin H, et al. Expanded 2-year follow-up of ranibizumab plus prompt or deferred laser or
369
Ophthalmology Retina Volume 3, Number 4, April 2019
26. 27. 28.
29.
triamcinolone plus prompt laser for diabetic macular edema. Ophthalmology. 2011;118:609e614. Elman MJ, Qin H, Aiello LP, et al. Intravitreal ranibizumab for diabetic macular edema with prompt versus deferred laser treatment. Ophthalmology. 2012;119:2312e2318. Regnier SA. How does drug coverage vary by insurance type? Analysis of drug formularies in the United States. Am J Manag Care. 2014;20:322e331. Ishibashi T, Li X, Koh A, et al. The REVEAL study: ranibizumab monotherapy or combined with laser versus laser monotherapy in Asian patients with diabetic macular edema. Ophthalmology. 2015;122:1402e1415. Korobelnik JF, Do DV, Schmidt-Erfurth U, et al. Intravitreal aflibercept for diabetic macular edema. Ophthalmology. 2014;121:2247e2254.
30. Wallick CJ, Hansen RN, Campbell J, et al. Comorbidity and health care resource use among commercially insured nonelderly patients with diabetic macular edema. Ophthalmic Surg Lasers Imaging Retin. 2015;46:744e751. 31. Kiss S, Chandwani HS, Cole AL, et al. Comorbidity and health care visit burden in working-age commercially insured patients with diabetic macular edema. Clin Ophthalmol. 2016;10:2443e2453. 32. Regillo CD, Callanan DG, Do DV, et al. Use of corticosteroids in the treatment of patients with diabetic macular edema who have a suboptimal response to anti-VEGF: recommendations of an expert panel. Ophthalmic Surg Lasers Imaging Retin. 2017;48:291e301. 33. Zhang J, Yu KF. What’s the relative risk? JAMA. 1998;280: 1690.
Footnotes and Financial Disclosures Originally received: October 24, 2018. Final revision: November 26, 2018. Accepted: November 30, 2018. Available online: December 7, 2018. Manuscript no. ORET_2018_423.
No animal subjects were included in this study. Author Contributions: Conception and design: Moulin, Chen, Burroughs, Vollman
1
Analysis and interpretation: Moulin, Adjei Boakye, Wirth, Vollman
2
Data collection: Moulin, Adjei Boakye, Chen Obtained funding: N/A
3
Overall responsibility: Moulin, Adjei Boakye, Wirth, Chen, Burroughs, Vollman
Saint Louis University Center for Health Outcomes Research (SLUCOR), St. Louis, Missouri. Department of Population Science and Policy, Southern Illinois University School of Medicine, Springfield, Illinois. Department of Internal Medicine, Southern Illinois University School of Medicine, Springfield, Illinois.
4
Department of Ophthalmology and Visual Sciences, Washington University School of Medicine, St. Louis, Missouri. Financial Disclosure(s): The author(s) have made the following disclosure(s): D.E.V.: Consultant – Verily Life Sciences, LLC.
HUMAN SUBJECTS: De-identified human subjects were included in this study. The Institutional Review Board of Saint Louis University approved this study. The study adhered to the tenets of the Declaration of Helsinki. Patient consent requirement was waived given we studied de-identified patient information.
370
Abbreviations and Acronyms: CI ¼ confidence interval; CPT ¼ current procedural terminology; DM ¼ diabetes mellitus; DME ¼ diabetic macular edema; FLP ¼ focal laser photocoagulation; HCPCS ¼ Healthcare Common Procedure Coding System; ICD-9-CM ¼ International Classification of Diseases, Ninth Edition, Clinical Modification; OR ¼ odds ratio; PPP ¼ procedures per patient; SD ¼ standard deviation; VEGF ¼ vascular endothelial growth factor. Correspondence: Thiago A. Moulin, MD, Saint Louis University, 3545 Lafayette Avenue, Salus Center, 4th Floor, Saint Louis, MO 63104. E-mail: tmoulin@gmail .com.