Statin Use and the Risk of Progression to Vision Threatening Diabetic Retinopathy

Abstract

Purpose:

This study aims to assess the effect of statins on progression from non-proliferative diabetic retinopathy (NPDR) to vision-threatening diabetic retinopathy (VTDR), proliferative diabetic retinopathy (PDR) or diabetic macular edema (DME).

Methods:

Two cohort studies using a US medical claims database from 2002–2019 including NPDR patients 18 years or older. A risk factor analysis performed a time-updating cox regression model assessing statin usage. A second new-user active comparator design analysis replicating a previously published study. Main outcomes included a new diagnosis of VTDR (composite of either PDR or DME) or DME and PDR individually for the risk factor study and included additional outcomes of new DR, NPDR, vitreous hemorrhage (VH) and tractional retinal detachment (TRD) for the new user study.

Results:

Risk factor analysis included 66,617 statin users with NPDR at baseline and 83,365 non-statin users. Of these, 27,325(18.2%) progressed to VTDR, 4,086(2.71%) progressed to PDR, and 22,750(15.1%) progressed to DME. After multivariable analysis, no protective effect of statin use was found for progression to VTDR, PDR, or DME (HR=1.01–3, p>0.33 for all comparisons). Replicated new user design analysis also showed no protective effect for statins on risk of development of DR(HR=1.03, 95%CI:0.99–1.07, p=0.13), PDR(HR=0.89, 95%CI:0.79–1.02, p=0.09), DME(HR=0.94, 95%CI:0.86–1.03, p=0.21), VH(HR=1.00, 95%CI:0.86–1.16, p=0.99), and TRD(HR=1.11, 95%CI:0.89–1.38, p=0.36).

Conclusion:

Statin use was found not to be protective for progression of DR regardless of study methodology. These results suggest that the specifics of the population studied rather than differing study methodology are important in assessing the effect of statins on DR progression.

Plain Language Summary:

In this study, we assess the effect of statins on progression of diabetic retinopathy (DR) from Non-proliferative Diabetic Retinopathy (NPDR) to vision-threatening diabetic retinopathy (VTDR), proliferative diabetic retinopathy (PDR) or diabetic macular edema (DME). The study found no protective effect for statins in the progression of DR. In addition, the findings of this study highlight the importance of considering the impact of how differing study populations may impact the results. Further studies are needed to better understand why analysis of statin use can lead to such contradictory findings across studies.

Introduction

Diabetic Retinopathy (DR) is the most common microvascular complication of diabetes mellitus (DM), with profound effects on the working age population.^1–3 Given the potential for visual impairment, and the increasing prevalence of disease,^4–7 much research has been undertaken to elucidate the effects of systemic medications and their potential for reducing vision loss in DR. Special attention has focused on examining the effects of cholesterol and cholesterol lowering medications on DR.^8–17

Statins have long been utilized to control blood cholesterol levels and decrease the risk of cardiovascular morbidity and mortality. Elevated blood lipid levels, especially cholesterol and low-density lipoprotein (LDL) have been associated with risk of progression to DR.^12–17 Naturally then, many studies have assessed the relationship between lipid control and progression to DR, with conflicting results. Some have demonstrated that lipid-lowering agents should be applied in prevention of DR in diabetes,^14,18–24 but others have failed to find a benefit.^23,25–29

Three large administrative database studies have demonstrated use of statins to decrease the progression of DR, including the incidence of initial DR, DME and PDR.^30–32 This finding has been supported by other, smaller studies.^33–35 However, other studies have suggested that statins are associated with reduced risk of developing DME, without any effect on vision loss and hard exudate aggravation.³¹ Currently there is no explicit statement indicating the use of lipid lowering drugs, specifically statins, in the treatment guidelines for DR.³⁶ Unfortunately, it is all too common a situation that when conflicting studies arise, there is not an attempt to assess whether the results occurred due to differing methodologies or differences in study populations. Therefore, further analysis is needed to understand why the relationship between statins and DR progression has been so inconsistent across the literature.

The purpose of this study is to assess if the use of statins is a protective factor in the progression of DR. To accomplish this, we initially conducted a study to assess if statins were a protective risk factor for progression of Non-Proliferative Diabetic Retinopathy (NPDR) to Vision Threatening Diabetic Retinopathy (VTDR), however upon seeing our results, we also elected to replicate a separate and distinct large administrative database study with a different underling population to better understand whether disparate results were secondary to methodological differences or the populations studied.

Methods:

Dataset

For both studies the Optum’s de-identified Clinformatics^® Data Mart Database was used. The database contains all outpatient medical claims (office visits, procedures, and medications given) as well as demographic data and some laboratory values for all patients enrolled in commercial and Medicare Advantage insurance plans obtained from a large US insurer The subset of data available for this study included all patients from January 1, 2002 to June 30, 2019. Due to the de-identified nature of the database, the University of Pennsylvania’s Institutional Review Board deemed this study exempt from review.

Statin use as a protective risk factor study

The methodology used to conduct this study closely mirrored a time updating study previously performed by this group and published elsewhere.³⁷ Briefly, a cohort of all patients who were diagnosed with NPDR based on ICD9 and ICD10 coding and had a history of diabetes mellitus was created. Previous studies of diabetic retinopathy diagnosis and treatment codes have been demonstrated to be valid for research purposes.^38–40 To account for the impact of the many lab values that have been shown to be associated with progression of DR, the index date was considered the first date that a patient had both a diagnosis of NPDR after at least one recorded laboratory value (hemoglobin A1c, hemoglobin or creatine for calculation of eGFR). Patients were excluded if at any time prior to the index date, they had a diagnosis of a more severe form of diabetic retinopathy, including VTDR. (See Supplemental Table 1 for all diagnosis, procedure, and drug codes used in the study). Statin use was assessed via prescription pharmacy claims data, which include the date of prescription fill and the days supplied. Patients were considered to be taking statins if they were within a period of “covered days” via the amount dispensed in each prescription.

Outcome Measures

The primary outcome was the progression to vision threatening diabetic retinopathy (VTDR) defined as a new occurrence of either DME or PDR after the index date. Secondary outcomes included DME and PDR individually. To avoid confusion between DME and cystoid macular edema, patients who had any form of intraocular surgery (e.g. vitrectomy, cataract surgery, trabeculectomy, etc.) were excluded from analysis for 120 days following surgery. This has been shown in database studies to increase the likelihood of truly incident disease (PMID: 23307227). Patients were allowed to return to the analysis after that time frame assuming they still met inclusion criteria. To increase the likelihood that these were incident cases of VTDR, any patient who met VTDR definition within two years of entry into the insurance plan was excluded from analysis. Of note, when analyzing the individual outcome, we censored patients who developed the other outcome before or on the date of outcome of interest. For example, when our outcome of interest is PDR, if a patient developed DME before PDR, this patient would be censored. Theoretically for patients who had both DME and PDR, if both diagnoses were on different dates, those patients would contribute to the total number of either PDR or DME, then the number of VTDR would be equal to the number of PDR plus number of DME. Therefore, the patients who had both PDR and DME on the same day were censored in the individual PDR or DME study but treated as having outcome in the composite (VTDR) study.

Statistical analysis and Covariates

Cox proportional hazard regression was performed. Censoring occurred if the patient left the plan, had any diagnosis of a disease or treatment of exclusion. Similarly, in considering the PDR and DME models separately, patients were censored from one once they met criteria for the other (i.e. censored in the PDR model if DME criteria were met). Covariates assessed included numerous demographics and systemic health conditions. To control for diabetes severity, the diabetes complications severity index (DCSI) was used. This score is created from six categories of diabetic complications using outpatient ICD9/ICD10 codes and applied to each patient. This metric has been validated and found to predict clinically relevant outcomes more accurately than traditional markers of diabetic severity such as glycosylated hemoglobin and duration of disease.⁴¹ By DCSI definition, all patients included in this study had a score of at least 1 due to their non-proliferative diabetic retinopathy status. As an additional severity factor, insulin use with prescription days covered was counted as time on/off the drug within the model. All variables capable of being assessed in a time-varying manner were updated every time new information was available. All analyses were conducted on a patient-level basis and not eye-level due to the lack of laterality in ICD9 coding which comprised a considerable portion of this study. P-values less than 0.05 were considered significant. Statistical analysis was performed using SAS (version 9.4; SAS Institute Inc., Cary, NC).

Replicated statin new-user cohort design study

Cohorts

This study was intended to replicate the methodology used by Kang et al.³⁰ This study differed in the previous study in that it focused on all diabetic patients who were over 40 and taking an oral antihyperglycemic (vs. only those with known NPDR). The earliest date that a patient met all 3 criteria was considered the index date. Exclusion occurred for anyone with a code for type 1 diabetes, having less than 1 year of data after the index date, having an LDL <100mg/dl, having a total cholesterol <160mg/dl, or any diagnosis or known treatment of VTDR or disease that could be confused with it prior to the index date. Patients with liver cirrhosis were also excluded as were any patients with the outcome in the first year after the index date. Cohorts were then created out of those who used statins and those who did not use statins. Statin users were not allowed to have had a previous statin prescription in the 3 months prior to the index date and had to have 80% prescription coverage over the first-year post index. The non-statin cohort was required to have zero statin use during the first-year post index date.

Outcomes

The outcome definition was also different in that it required 3 total diagnosis codes of an outcome to count. In addition, the outcomes were divided differently in that any new diabetic retinopathy, NPDR, PDR, DME, vitreous hemorrhage (VH) and tractional retinal detachment (TRD) were all treated as individual outcomes. Due to a difference in the databases used (The Kang study had complete follow up until death vs the Optum database which patients drop out once they change insurance), a sensitivity analysis was performed that only required a single diagnosis of the above to count as an outcome.

Covariates and statistical analysis

Covariates that were assessed included age, sex, hypertension (defined by either ICD code or use of an HTN medication at index), ischemic heart disease, chronic kidney disease, dialysis, heart failure, ischemic stroke, peripheral arterial disease, Charlson comorbidity index and diabetic severity coded as diabetic neuropathy, diabetic foot disease hx of amputation or none of the above. An inverse proportional treatment weighting was performed based on the likelihood of being prescribed a statin medication. Cox proportional hazards regression models were then analyzed for each outcome, controlling for all covariates.

Results

Risk Factor study

150,252 patients with NPDR diagnosis met inclusion and exclusion criteria for this analysis (Table 1/Figure 1). Of these patients, 27,325 (18.2%) progressed to VTDR, including 4,086 (2.71%) and 22,750 (15.1%) progressed to PDR and DME, respectively. The mean age of patients was 64.3 (SD 13.2) years old among non-statin users and 67.4 (SD 10.5) years old among statin users (Table 1). 49.9% of non-statin users and 53.1% of statin users were male (Table 1). Of non-statin users and statin users, respectively, 46.9% and 46.2% were White, 15.5%, and 13% were Black, 4.5% and 5.0% were Asian, and 17.7% and 19.0% were Hispanic (Table 1). The mean hemoglobin a1c was 7.4 (SD 2.1) among statin users and 7.3 (mean 2.7) among non-statin users. 23.7% of statin users used insulin, compared to 17.4% of non-statin users (Table 1). Among statin users, the mean hemoglobin a1c was

Table 1:

Baseline Characteristics stratified by Statin Use^**

	No Use of Statin at Baseline (N=83635)	Use of Statin at Baseline (N=66617)	p value
Age			<0.001
Mean (SD)	64.3 (13.2)	67.4 (10.5)
Gender			<0.001
Female	41889 (50.1%)	31212 (46.9%)
Male	41746 (49.9%)	35405 (53.1%)
Race			<0.001
White	39261 (46.9%)	30786 (46.2%)
Black	12934 (15.5%)	8638 (13.0%)
Asian	3728 (4.5%)	3343 (5.0%)
Hispanic	14772 (17.7%)	12650 (19.0%)
Unknown	12940 (15.5%)	11200 (16.8%)
History of hypercholesterolemia	69383 (83.0%)	63147 (94.8%)	<0.001
DCSI *			<0.001
Mean (SD)	4.0 (2.6)	4.5 (2.6)
1	18752 (22.4%)	10676 (16.0%)
[2, 3]	23002 (27.5%)	16837 (25.3%)
4	10361 (12.4%)	8382 (12.6%)
[5, 12]	31520 (37.7%)	30722 (46.1%)
Hemoglobin A1c			0.02
N	66874	54445
Mean (SD)	7.3 (2.7)	7.4 (2.1)
Use of insulin			<0.001
No	69060 (82.6%)	50825 (76.3%)
Yes	14575 (17.4%)	15792 (23.7%)

^**Comprehensive baseline and demographic characteristics as previously reported with novel statin information reported here.³⁷

^*Diabetes complications severity index

Figure 1.

Flowchart of Patients with NPDR That Met Inclusion and Exclusion Criteria

After controlling for all covariates, cox model analysis showed statins to not be associated with any significant protective effect for progression to VTDR (HR 1.01, 95% CI: 0.99, 1.04, p = 0.41) (Table 2). After analyzing PDR and DME individually, no protective effect was seen with use of oral statin for progression to PDR (HR 1.03, 95% CI: 0.97, 1.10, p=0.33) or progression to DME (HR 1.01, 95% CI: 0.99, 1.04, p=0.38) (Table 2). Of note, a sensitivity analysis requiring only 1 diagnosis code demonstrated no difference in outcomes compared to our primary analysis.

Table 2.

Multivariate cox regression results for progression to VTDR, PDR, and DME^*

		Multivariate Cox Model for progression to VTDR		Multivariate Cox Model for progression to PDR		Multivariate Cox Model for progression to DME
Variable	Level	HR (95% CI)	P-value	HR (95% CI)	P-value	HR (95% CI)	P-value
Age		0.993 (0.992, 0.995)	<0.001	0.982 (0.979, 0.985)	<0.001	0.996 (0.995, 0.998)	<0.001
Gender (female ref)	Male	0.93 (0.90, 0.95)	<0.001	1.08 (1.02, 1.15)	0.02	0.90 (0.87, 0.92)	<0.001
Race (White ref)	Unknown	0.95 (0.88, 1.03)	<0.001	0.96 (0.78, 1.18)	<0.001	0.94 (0.87, 1.03)	<0.001
	Hispanic	0.96 (0.93, 1.00)		1.20 (1.10, 1.31)		0.92 (0.89, 0.96)
	Asian	0.81 (0.76, 0.87)		0.86 (0.72, 1.02)		0.80 (0.74, 0.86)
	Black	0.91 (0.88, 0.95)		1.05 (0.95, 1.15)		0.89 (0.86, 0.93)
DCSI (1 ref)	[5, 12]	1.67 (1.59, 1.75)	<0.001	2.25 (1.97, 2.57)	<0.001	1.56 (1.48, 1.65)	<0.001
	4	1.49 (1.41, 1.57)		1.71 (1.48, 1.97)		1.44 (1.36, 1.53)
	[2, 3]	1.24 (1.18, 1.30)		1.38 (1.22, 1.58)		1.21 (1.15, 1.28)
Hemoglobin A1c		1.04 (1.03, 1.05)	<0.001	1.10 (1.09, 1.12)	<0.001	1.02 (1.02, 1.03)	<0.001
Hemoglobin A1c Unknown		1.41 (1.33, 1.50)	<0.001	2.06 (1.78, 2.38)	<0.001	1.26 (1.18, 1.35)	<0.001
Use of oral statin	Yes	1.01 (0.99, 1.04)	0.41	1.03 (0.97, 1.10)	0.33	1.01 (0.99, 1.04)	0.38

^*In multivariate analysis, use of fenofibrate was adjusted for in order to isolate the cholesterol lowering effect to statins

Replication study

To determine whether the differences in results between our analysis and those found in a previously published analysis³⁰ were due to methodological differences or differences in populations, we also performed an analysis of our data using their methodology. In this replication analysis, 440,143 patients were included. 196,926 (44.7%) were female, with mean age of 65.6 (SD 10.3) years old (Table 3). 152,959 (34.8%) were statin-users (Table 3). 8,217 (1.9%) had NPDR, 11,554 (2.6%) developed DR, 1,042 (0.2%) developed PDR, 2,115 (0.5%) developed DME, 775 (0.2%) developed vitreous hemorrhage, and 350 (0.1%) developed Tractional Retinal Detachment (TRD) (Table 3). After controlling for all covariates, cox model showed use of statins was not associated with risk of development of DR (HR = 1.03, 95% CI: .99, 1.07, p =0.13), PDR (HR = 0.89, 95% CI: 0.79, 1.02, p=0.09), DME (HR = 0.94, 95% CI: 0.86, 1.03, p-value = 0.21), vitreous hemorrhage (HR = 1.00, 95% CI: 0.86, 1.16, p=0.99), or formation of a tractional retinal detachment (HR = 1.11, 95% CI: 0.89, 1.38, p = 0.36). A slight increased risk of progression to NPDR was noted with statin use (HR = 1.06, 95% CI: 1.01, 1.11, p=0.02), (Table 4).

Table 3:

Baseline Characteristics and Descriptive Statistics of Outcome Measures of population replicating Kang et al. 2019 methodology³⁰

	Before Matching			After Matching
Variable	Non-Statin User (N = 318813)	Statin User (N = 152957)	SMD	Non-Statin User (N = 287186)	Statin User (N = 152957)	SMD
Gender			0.036			0.004
Female	147928 (46.4%)	68256 (44.6%)		128670 (44.8%)	69256 (44.6%)
Male	170885 (53.6%)	84701 (55.4%)		158516 (55.2%)	84701 (55.4%)
Age at baseline Mean (SD)	65.77 (10.42)	65.91 (10.05)	0.014	65.4 (10.4)	65.9 (10.1)	0.047
Office Visits Mean (SD)	3.71 (4.19)	3.99 (4.37)	0.066	3.77 (4.23)	3.99 (4.37)	0.053
Hypertension diagnosis with HTN drug use at baseline	88559 (27.8)	44461 (29.1)	0.029	78925 (27.5%)	44461 (29.1%)	0.035
Ischemic heart disease at baseline	35398 (11.1)	29978 (19.6)	0.237	35379 (12.3%)	29978 (19.6%)	0.200
Chronic kidney disease at baseline	28679 (9.0)	15168 (9.9)	0.031	25368 (8.8%)	15168 (9.9%)	0.037
Ischemic stroke at baseline	14601 (4.6)	10215 (6.7)	0.091	14421 (5.0%)	10215 (6.7%)	0.071
Heart failure at baseline	14276 (4.5)	9710 (6.3)	0.083	13628 (4.7%)	9710 (6.3%)	0.070
Peripheral arterial disease at baseline	17666 (5.5)	10233 (6.7)	0.048	15741 (5.5%)	10233 (6.7%)	0.051
Severity of diabetic disease at baseline			0.017			0.032
No disease	288826 (90.6)	138263 (90.4)		262124 (91.3%)	138263 (90.4%)
Diabetic neuropathy	23315 (7.3)	11129 (7.3)		19338 (6.7%)	11129 (7.3%)
Diabetic foot	6029 (1.9)	3190 (2.1)		5135 (1.8%)	3190 (2.1%)
Amputation	643 (0.2)	375 (0.2)		589 (0.2%)	375 (0.2%)
Charleson Comorbidity Index at baseline
Mean (SD)	1.79 (1.86)	1.86 (2.03)	0.033	1.73 (1.84%)	1.86 (2.03%)	0.067
Outcome Measures
DR	24961 (7.8)	14707 (9.6)	0.063	24961 (8.7%)	14707 (9.6%)	0.032
NPDR	18416 ( 5.8)	10800 (7.1)	0.052	18416 (6.4%)	10800 (7.1%)	0.026
Proliferative DR	1656 (0.5)	871 (0.6)	0.007	1656 (0.6%)	871 (0.6%)	0.001
DME	3669 (1.2)	2061 (1.3)	0.018	3669 (1.3%)	2061 (1.3%)	0.006
Vitreous heme	1493 (0.5)	849 (0.6)	0.012	1493 (0.5%)	849 (0.6%)	0.005
TRD	588 (0.2)	343 (0.2)	0.009	588 (0.2%)	343 (0.2%)	0.004

Table 4.

Multivariable Cox Regression Analysis with time-constant covariates^* for Diabetic Retinopathy, NPDR, PDR, DME, Vitreous Heme, TRD replicating Kang et al. 2019 methodology³⁰

		Diabetic Retinopathy		NPDR		PDR		DME		Vitreous Heme		TRD
Variable		Hazard Ratio (95%CI)	P-value	Hazard Ratio (95%CI)	P-value	Hazard Ratio (95%CI)	P-value	Hazard Ratio (95%CI)	P-value	Hazard Ratio (95%CI)	P-value	Hazard Ratio (95%CI)	P-value
Age		1.00 (1.00, 1.00)	0.02	0.99 (0.99, 1.00)	<0.001	0.96 (0.95, 0.96)	<0.001	0.99 (0.98, 0.99)	<0.001	0.98 (0.97, 0.99)	<0.001	0.97 (0.96, 0.98)	<0.001
Gender (Male)		0.92 (0.89, 0.96)	<0.001	0.96 (0.92, 1.00)	0.06	1.02 (0.90, 1.16)	0.71	0.97 (0.89, 1.06)	0.47	1.20 (1.04, 1.39)	0.02	1.94 (1.53, 2.45)	<0.001
Number of Office visits		0.96 (0.96, 0.97)	<0.001	0.96 (0.95, 0.96)	<0.001	0.91 (0.89, 0.93)	<0.001	0.95 (0.94, 0.96)	<0.001	0.99 (0.97, 1.00)	0.08	1.00 (0.97, 1.03)	0.99
HTN		1.03 (0.99, 1.07)	0.10	1.00 (0.96, 1.05)	0.99	0.94 (0.83, 1.07)	0.34	0.99 (0.90, 1.08)	0.76	0.96 (0.83, 1.11)	0.59	0.94 (0.75, 1.16)	0.54
Ischemic heart disease		0.85 (0.81, 0.88)	<0.001	0.89 (0.85, 0.94)	<0.001	1.00 (0.87, 1.15)	0.99	0.86 (0.77, 0.94)	0.002	1.05 (0.89, 1.24)	0.54	0.86 (0.67, 1.11)	0.25
Chronic kidney disease		1.17 (1.11, 1.23)	<0.001	1.12 (1.05, 1.19)	<0.001	1.45 (1.22, 1.72)	<0.001	1.16 (1.03, 1.30)	0.01	1.15 (0.93, 1.41)	0.19	1.06 (0.76, 1.47)	0.74
Dialysis		0.71 (0.67, 0.75)	<0.001	0.73 (0.69, 0.79)	<0.001	0.96 (0.80, 1.16)	0.70	0.72 (0.63, 0.83)	<0.001	0.91 (0.72, 1.15)	0.43	0.84 (0.56, 1.26)	0.40
Ischemic stroke		0.72 (0.69, 0.76)	<0.001	0.79 (0.75, 0.84)	<0.001	0.82 (0.69, 0.97)	0.02	0.87 (0.78, 0.98)	0.02	0.85 (0.70, 1.03)	0.11	1.13 (0.84, 1.52)	0.43
Heart failure		0.86 (0.82, 0.91)	<0.001	0.87 (0.82, 0.92)	<0.001	1.13 (0.96, 1.34)	0.15	0.92 (0.81, 1.04)	0.18	1.27 (1.05, 1.54)	0.01	0.99 (0.70, 1.40)	0.96
Peripheral arterial disease		0.88 (0.84, 0.92)	<0.001	0.83 (0.79, 0.88)	<0.001	1.05 (0.89, 1.22)	0.58	0.81 (0.72, 0.91)	<0.001	0.91 (0.76, 1.10)	0.35	0.98 (0.72, 1.33)	0.91
Severity of diabetic disease
	Amputation	3.09 (2.69, 3.56)	<0.001	2.93 (2.47, 3.48)	<0.001	7.43 (5.49, 10.05)	<0.001	4.14 (3.11, 5.52)	<0.001	4.42 (2.89, 6.75)	<0.001	2.26 (0.92, 5.56)	0.27
	Diabetic foot	1.63 (1.53, 1.73)		1.53 (1.42, 1.64)		1.72 (1.40, 2.11)		1.48 (1.28, 1.72)		1.30 (1.02, 1.67)		1.04 (0.68, 1.59)
	Diabetic neuropathy	1.63 (1.56, 1.70)		1.62 (1.54, 1.71)		1.60 (1.38, 1.86)		1.58 (1.43, 1.75)		1.13 (0.94, 1.36)		0.90 (0.67, 1.22)
	No disease	Ref.		Ref.		Ref.		Ref.		Ref.		Ref.
Charleson Comorbidity Index		1.18 (1.17, 1.19)	<0.001	1.17 (1.16, 1.18)	<0.001	1.15 (1.12, 1.18)	<0.001	1.18 (1.16, 1.20)	<0.001	1.12 (1.08, 1.15)	<0.001	1.03 (0.97, 1.09)	0.31
Statin		1.03 (0.99, 1.07)	0.13	1.06 (1.01, 1.11)	0.02	0.89 (0.79, 1.02)	0.09	0.94 (0.86, 1.03)	0.21	1.00 (0.86, 1.16)	0.99	1.11 (0.89, 1.38)	0.36

^*covariates controlled for including age, gender, statin use, office visits, hypertension diagnosis in addition to drug use, ischemic heart disease, chronic heart disease, dialysis, ischemic stroke, heart failure, peripheral arterial disease, severity of diabetic disease (amputation, diabetic foot, diabetic neuropathy, no disease, charleson comorbidity index)

Discussion

In this study, we sought to assess the association between statin use and progression of DR to its vision threatening forms. In our primary analysis, we found that statin use was not associated with a lower risk of progression to VTDR or either of its component diseases PDR or DME. Once our results were found to be contradictory with other reports in similar types of data, a replication analysis of a different study was performed to better understand the etiology of our conflicting findings.^19,30–35 These replication analyses also found no protective benefit of statin use with regards to progression to DR, PDR, DME, vitreous hemorrhage, or TRD.³⁰

In contrasting our risk factor results with Kang et al., the study we replicated, we were concerned that some of the methodological choices that differed between the studies had impacted the results.³⁰ Notably, Kang et al. extracted information on all diabetic patients taking oral diabetic medications (and by definition, excluding type 1 diabetics) regardless of retinopathy status, whereas our study focused on NPDR patients specifically (independent of type 1 or type 2 DM status). Kang et al. also assessed overall health state utilizing the Charleston comorbidity index which is primarily insightful for inpatient rather than outpatient populations.⁴² Similarly, the severity of diabetes was assessed only through diagnosed neuropathy, foot ulcer disease, amputation and DM medication usage. In contrast, in our risk factor analysis, we used the DCSI – a measure of severity of diabetes more appropriate to the outpatient setting,⁴¹ hemoglobinA1c level and insulin use. Despite these differences, when we modeled a study based on Kang et al’s methodologic choices, we still failed to find a protective effect for statins in our population. This along with the relatively similar baseline health characteristics between the two studies, strongly points to genetic or environmental differences between the populations that may have led to the disparate results.

Another possibility for the conflicting results, and one that often complicates assessments of statins in health care databases is the “healthy user effect” frequently seen in patients who use statins. This refers to the phenomenon whereby statins have been shown to reduce incidence of events that cannot be conceivably attributed to statins (i.e. car crashes, emergency room visits, etc.).⁴³ In another recent report looking at a Japanese medical claims database, statin use was associated with decreased risk of incidence of DR, DME, any treatment for DR, laser photocoagulation and vitrectomy.³² Although possible genetic effects or environmental effects may have impacted this study as well, it is also conceivable a healthy user effect was seen. Comparing “health status” across studies is difficult because the authors did not control for many of the systemic health conditions assessed in our or Kang et al.’s, nor did they control for underlying diabetic severity beyond duration of disease. One argument for a healthy user effect occurring in the Japanese population was that the author’s found insulin use to have had the lowest odds of DR progression (much stronger than statins), the opposite of what would be expected given the insulin is typically a last line of therapy used in DM care.

More difficult to explain are why our results conflicted with another recent study that assessed a United States medical claims database similar, but distinct, from the one used in our analyses. Consistent with the other two positive results studies, this U.S. study also reported that statins users were less likely to develop NPDR, VTDR, and also showed they were less likely to receive intravitreal injections of anti-VEGF medication, laser treatments, or vitrectomy.³¹ Superficially since both studies were done using data from the US it would seem that the populations assessed were comparable, however when looking at the baseline characteristics of each study, the average age of our study population was nearly 20 years older. Again, health characteristics are difficult to compare as they were not published for the comparison U.S. study, but given the differences in age and the dramatically lower percentage of outcomes that occurred (8.7–9.6% in our study vs. <1.0% in their study), it is possible population health effects differentially impacted the results.

While this study has many strengths, we do acknowledge several limitations to this study. First, the de-identified medical claims data used in this study does not allow for further investigation or referencing of clinical and laboratory data. However, the ICD and CPT codes for diabetic retinopathy and its treatments have been shown to have good reliability.^38–40 In addition, this data is derived from a single medical claims database and as has been demonstrated above, may not represent the results of other databases with different populations across the globe, and even in the US (eg Medicare or Medicaid databases). Next, while we assume patients who were prescribed statins were taking the statins, we are unable to confirm that the statins were taken as prescribed. Furthermore, due to the nature of the database we are unable to definitely determine duration of diabetic disease and type of diabetes (1 or 2). While both are known associations for progression of diabetic retinopathy, we have attempted to reduce this limitation by controlling for the diabetes complications severity index and insulin use. However, we are unable to rule out the possibility that residual unmeasured confounding could exist due to a lack of these variables. Lastly, although we censored patients after eye surgery for 120 days, we cannot rule out the possibility that a very small number of patients would have newly diagnosed CME which could have falsely increased the rate of DME diagnoses. However, there is no reason to believe this would have impacted statin users differentially from non-statin users.

The study found no protective effect for statins in the progression of DR. In addition, the findings of this study highlight the importance of considering the impact differing study populations and environmental effects may impact the results. Further studies are needed to better understand why analysis of statin use can lead to such contradictory findings across studies.

Key Points:

• The literature demonstrates conflicting results on the effect of statins on progression of non-proliferative diabetic retinopathy to vision threatening diabetic retinopathy and, on the factors, responsible for the literature’s conflicting results

• In this retrospective cohort study, statin use was not associated with decreased progression to diabetic macular edema or proliferative diabetic retinopathy.

• Replication analysis of a previous study using differing methodology demonstrated similar negative results.

• It is important to consider the specifics of the population studied in extrapolating results on the effect of statins on progression of DR.