Statin Therapy and Risk of Polyneuropathy in Type 2 Diabetes: A Danish Cohort Study

Diabetic polyneuropathy (DPN) affects up to 50% of all patients with diabetes and is associated with an increased risk of foot ulcers and amputations (1,2). Metabolic syndrome components, including obesity, hypertension, hypertriglyceridemia, and low levels of HDL cholesterol are highly prevalent in patients with newly diagnosed type 2 diabetes (3). These components have been associated with an increased risk of DPN (4–6), possibly through reactive oxygen species, local nerve inflammation, and impaired endoneurial capillary function (1,2,7,8).

Statins may have the potential to prevent DPN in patients with type 2 diabetes through lipid-lowering and pleiotropic functions, the latter including endothelial activation and anti-inflammatory and antioxidative effects (9,10). Findings from observational studies in type 2 diabetes populations have supported this hypothesis, reporting a 15–35% risk reduction of diabetic neuropathy with prevalent statin use (11–13). In contrast, several small (N < 50) clinical trials of patients with prevalent DPN were unable to show any beneficial effect of statin therapy on DPN improvement compared with placebo (14,15).

Despite theories of any statin benefit, animal studies have suggested that statins may be neurotoxic, and several case reports and observational studies of individuals without diabetes have associated statin initiation with neuropathy symptoms (16–18). Thus, the impact of statin use on DPN risk remains controversial.

Previous observational studies of statin use and neuropathy risk in type 2 diabetes may have been biased toward a beneficial statin effect (e.g., by misclassification of a statin-induced reduction in cerebrovascular events as reduced DPN risk) (19). Also, focusing on prevalent statin users may have increased the risk of healthy adherer bias and prevented the identification of potential acute adverse effects of statin treatment (20). We therefore conducted a large cohort study of patients with incident type 2 diabetes in Denmark, applying a new-user design and using recently validated diagnosis codes for hospital-diagnosed DPN (19), to clarify whether statin treatment has an impact on the risk of developing DPN.

The Danish National Health Service provides health care services free of charge to all Danish residents, including access to hospitals and general practitioners (21). We linked individual-level data from population-based Danish medical registries using the unique central personal registration number assigned to all Danish residents (22). The Danish National Patient Registry (DNPR) provided complete hospital contact history (23), and the Danish National Prescription Registry (NPR) provided information on all prescribed medications redeemed at community pharmacies in Denmark (24). Statins are not sold over the counter in Denmark, and prescriptions for these drugs are fully captured in the NPR (24). For a subgroup analysis, laboratory lipid tests were ascertained from the clinical laboratory information system research database, which contains clinical biochemistry data from primary and secondary health care visits for the central and northern Danish regions (1.8 million residents) (25).

We used the DNPR and the NPR to establish a study population of patients with incident type 2 diabetes. The type 2 diabetes study population consisted of all people resident in Denmark aged ≥30 years with either 1) a first-time occurrence of an inpatient or outpatient hospital discharge diagnosis of diabetes or 2) a first-time filled prescription for a glucose-lowering drug issued by either a primary care or a hospital-based physician between 2 January 2002 and 5 July 2016, thus including both patients with diabetes managed in the community in primary care and patients with diabetes with hospital contact. In an earlier validation study, this algorithm for identifying diabetes was found to have a positive predictive value (PPV) of 97% for hospital-based diagnoses and 95% for prescription-based diagnoses (26). Patients who were <30 years old at the time of their first record of diabetes were excluded to limit the number of patients with type 1 diabetes. We further excluded those with a previous diagnosis of DPN (algorithm defined below), those with a diagnosis of other polyneuropathies or disorders of the peripheral nervous system (G60–G64), and those with former statin use (see definition below) before the index date (Supplementary Figs. 1 and 2).

We obtained complete information on statin use from the NPR. A statin exposure assessment window was established to categorize the patients into new users, prevalent users, and nonusers of statins (Supplementary Fig. 2). The statin exposure assessment window extended from 180 days before to 180 days after the first incident diabetes record to allow time for a diagnostic work-up, blood lipid testing, and decisions surrounding the advisability of starting statin therapy. New statin users were defined as filling their first-ever statin prescription within the 360-day exposure assessment window (Supplementary Fig. 2). Prevalent statin users were defined as also filling at least one statin prescription within this window but also had filled their first-ever statin prescription before this window. Nonusers had not filled a statin prescription since 1995 up until the end of the exposure assessment window. We also identified and excluded from further study a small group of former users, defined as having filled one or more statin prescriptions before the 360-day exposure assessment window but not during that period. Follow-up was initiated at the end of the exposure assessment window, defined as the index date, corresponding to 180 days after the first record of diabetes (Supplementary Fig. 2).

We used a DPN algorithm (described below) that was based on ICD-10 diagnosis codes. Before conducting this study, we validated our DPN algorithm with information from medical record review. In the validation study, we first identified patients with type 2 diabetes with hospital diagnosis codes indicative of DPN. Next, we randomly selected a validation sample, collected their medical records, and performed a medical record review. DPN was verified by either a positive nerve conduction test, the presence of one or more prespecified symptoms of DPN (e.g., numbness, paresthesia), the presence of one or more prespecified objective signs of DPN (e.g., abnormal vibration, abnormal light touch, abnormal pinprick), or physician notes documenting the presence of DPN (19). The algorithm showed a positive predictive value of 74% for hospital-diagnosed DPN in the DNPR (n = 79) assessed as either 1) a primary or secondary discharge diagnosis of polyneuropathy, unspecified (G62.9), or diabetic polyneuropathy (G63.2) or 2) a primary discharge diagnosis of diabetes with neurological complications (E10.4–14.4) (19).

We searched the DNPR and the NPR for covariates associated with statin therapy and DPN. We obtained information on biological sex, age, diagnoses of hyperlipidemia, obesity, and hypertension; use of other lipid-lowering drugs; antihypertensive drug use; insulin therapy; macrovascular complications; microvascular complications; and smoking-related disorders (1,4–6). As other possible causes of neuropathy symptoms, we also included data on alcohol-related disorders, HIV/AIDS, cancer, chemotherapy treatment, hypothyroidism, B12 and other B-vitamin deficiencies, connective tissue disease, and neuropathy-related infections as covariates (27). As well, we included chronic pulmonary disease, gastrointestinal and liver disease, and dementia since these covariates could influence the choice to initiate statin therapy. As a measure of frailty, we included the number of inpatient hospitalizations and inpatient hospitalization days within 1 year before the index date. Codes and definitions are provided in Supplementary Table 1.

We first compiled descriptive statistics for study participants as of the index date. We followed all patients from their index date until occurrence of first-time hospital-diagnosed DPN, death, emigration, or study end (1 January 2018), whichever came first. We plotted crude cumulative incidence curves for DPN stratified by statin use, treating death as a competing risk. In an intention-to-treat analysis, we used Cox proportional hazards regression to compute crude and adjusted hazard ratios (aHRs) with 95% CIs, comparing new and prevalent users with statin nonusers. The underlying time scale was years after the index date. The adjusted Cox regression analysis included age (continuous variable), sex, index year (continuous variable), insulin use, diagnosis-defined hyperlipidemia and obesity, macrovascular complications, other microvascular complications, hypertension, smoking-related disorders, and alcohol-related disorders (1,4–6). The assumption of proportional hazards was verified by visual inspection of log-minus-log plots. We stratified the analyses by sex and age-groups (30–50 years, 50–70 years, and >70 years) to investigate the potential effect measure modification. Finally, we stratified the follow-up period into 1-year intervals to reveal any acute nerve toxicity effect or potential protopathic bias (i.e., early symptoms of yet-undiagnosed DPN that trigger statin initiation).

We performed nine additional analyses:

1. An extended adjusted analysis, including all covariates shown in Table 1 

2. A propensity score–matched analysis to optimize balancing of confounding factors

3. An analysis with additional adjustment for baseline LDL cholesterol (LDL-C) levels, triglyceride levels, and HDL cholesterol levels in a geographical subpopulation linkable to the clinical laboratory information system research database

4. An on-treatment analysis to account for possible misclassification of statin use during follow-up

5. An analysis that delayed the start of follow-up by 1 year to account for protopathic bias (i.e., early symptoms of yet-undiagnosed DPN that trigger statin initiation)

6. Categorization of new users of simvastatin and atorvastatin into low-dose (<40 mg) and high-dose (≥40 mg) users on the basis of the dose of their initial prescription

7. An analysis using the neuropathy definition provided in the article by Nielsen and Nordestgaard (12)

8. A bias analysis performed using the website https://www.clepan.com, which assumed a DPN misclassification of 26% (diagnostic PPV 74% [19]) and completeness of polyneuropathy diagnoses of 20% (28)

9. An analysis using all-cause mortality as an outcome in which statins have a documented effect as a positive control for our main analysis

We used Stata 14.0 statistical software (StataCorp) for all analyses. According to Danish law, approval from an ethics committee was not required.

Between 2 January 2002 and 5 July 2016, we identified a total of 310,676 patients newly diagnosed with diabetes. Of these, 292,139 (94%) were living in Denmark and alive on the index date. After applying exclusion criteria, 259,625 patients remained, among whom 59,255 (22.8%) were new statin users, 75,528 (29.0%) were prevalent users, and 124,842 (48.1%) were nonusers (Supplementary Fig. 1).

Table 1 presents descriptive data for each group. Males represented 57.6% of all statin users. Prevalent statin users had a higher median age (67 years [IQR 60–74]) than new users (60 years [52–68]) and nonusers (59 years [47–71]). The median time from first statin prescription until the index date was 180 days (138–180) for new statin users and 1,823 days (1,000–2,986) for prevalent statin users.

During follow-up, we identified 6,677 patients with incident hospital-diagnosed DPN, 60,628 patients who died, and 1,606 patients who emigrated. Median follow-up time for the total cohort was 6.21 years (IQR 3.39–9.57) and by exposure group was 6.75 years (4.08–9.79) for new users, 5.58 years (3.07–8.04) for prevalent users, and 6.52 years (3.30–10.50) for nonusers. Corresponding incidence rates of DPN per 1,000 person-years were 4.0 (95% CI 3.8–4.2), 3.8 (3.6–3.9), and 3.8 (3.7–4.0), respectively. The cumulative incidences of DPN were similar for the three exposure groups (Fig. 1).

The aHRs for hospital-diagnosed DPN were 1.05 (95% CI 0.98–1.11) for new statin users and 0.97 (0.91–1.04) for prevalent users (Table 2). While new users were at slightly increased risk of DPN during the first year of follow-up (1.31 [1.12–1.53]), this vanished after ≥2 years of follow-up (Fig. 2). Biological sex appeared to modify the association. Thus, mainly female new users (1.20 [1.08–1.33]) and female prevalent users (1.11 [0.99–1.25]) were at increased risk of DPN (Table 2). Risk estimates slightly >1 also were observed among new statin users aged 30–50 years (1.21 [1.06–1.39]) and prevalent users (1.22 [0.98–1.51]) (Table 2).

Additional adjustment for all variables shown in Table 1 did not materially change the association (Supplementary Table 2). Using a propensity score–matched population (N = 91,922 [35.4%]) yielded similar results to those of the main analysis for new users (HR 1.02 [95% CI 0.93–1.12]) (Supplementary Table 3). Additional adjustment for baseline lipid levels in a geographic subpopulation (n = 55,176 [21.3%]) led the association toward the null for new statin users (aHR 0.98 [95% CI 0.84–1.14]), even though new statin users had lowered their LDL-C levels by >40% from baseline up to 1 year of follow-up (Supplementary Tables 4 and 5).

During follow-up, 45% of the statin nonusers initiated statin therapy with a median time from index date until first statin prescription of 666 days (IQR 246–1,442). Concurrently, 39% of the new users and 26% of the prevalent users apparently discontinued statin therapy, with a median time from index date until first statin discontinuation of 930 days (463–1,780) and 990 days (519–1,760), respectively.

Additional censoring of patients when they either initiated or discontinued statin therapy (on-treatment analysis) yielded an aHR of 1.17 (95% CI 1.09–1.27) for new statin users and 1.06 (0.98–1.16) for prevalent users (Supplementary Table 6). This slightly increased risk for new statin use was mainly driven by the first year of follow-up (1.42 [1.21–1.68]) (Supplementary Fig. 3). When we investigated the impact of initial statin dose, high-dose initiation was not associated with a higher risk (1.00 [0.93–1.08]) compared with low-dose initiation (1.11 [1.02–1.21]). The slightly increased low-dose risk estimate was mainly driven by higher risk in the first year of follow-up (Supplementary Fig. 4). In our bias analysis, which investigated the impacts of up to a 26% misclassification rate and 20% rate of completeness of the outcome algorithm, the estimate did not substantially change (risk ratio 1.00, data not shown). When we used the outcome definition of Nielsen and Nordestgaard (12), prevalent statin use was not associated with diabetic neuropathy (0.96 [0.89–1.03]). Delaying the start of follow-up to 1 year after the index date moved the estimate toward the null for new statin users (1.01 [0.95–1.08]) and left the estimate unchanged for prevalent statin users (0.98 [0.91–1.06]). Finally, as a positive control, we found that all-cause mortality was 25% decreased in both new and prevalent statin users, consistent with previous studies (29) (Supplementary Table 7).

In this large cohort study, we found no association between statin therapy and hospital-diagnosed DPN risk in our main analysis. Findings were similar in propensity score–matched analyses and with additional adjustment for pretreatment blood lipid levels. A modestly increased risk of DPN was observed in on-treatment analyses and during the first year of follow-up. Overall, our results indicate that statin therapy is unlikely to cause or mitigate DPN risk, although a small acute risk of harm cannot be excluded.

In contrast to our findings, the Australian Fremantle cohort study found a 35% reduction in screen-detected diabetic neuropathy risk among prevalent statin users (11). Similarly, Danish and Taiwanese cohort studies reported a 34% and 15% reduced risk of diabetic neuropathy, respectively, with prevalent statin use (12,13). There are several possible explanations for these divergent findings. First, the three previous studies restricted statin exposure to prevalent use. The fact that prevalent versus new use of statins was in general associated with lower DPN risk in our study may point to healthy statin adherer bias (20). Second, in contrast to Nielsen and Nordestgaard (12) and Kang et al. (13), we applied a validated outcome algorithm to increase the likelihood of including only true DPN and no other types of neuropathy (e.g., mononeuropathy) (19). However, changing our outcome definition to the codes used by Nielsen and Nordestgaard did not materially change the association for prevalent users. Nielsen and Nordestgaard initiated follow-up on the date of the patients’ first diabetes record. This may have led to misclassification of nonusers because we found that 80% of all new users initiated statin therapy within 180 days after their first record of diabetes (data not shown). Because our data indicated a modestly increased early risk of DPN among new statin users, any misclassification of new users as nonusers may create a false impression of a protective association when prevalent users are compared with nonusers.

The time-varying risk of hospital-diagnosed DPN for new statin users may be explained by protopathic bias. Early signs of DPN may trigger a contact with the health care system and then initiation of statin therapy at the same occasion. Because of diagnostic work-up time, the diagnosis of DPN may have been first recorded at a later time. Also, some DPN events may represent atypical forms of neuropathy induced by aggressive antidiabetic treatment rather than by statin therapy (30). An alternative explanation may be pathophysiological (i.e., an acute neurotoxicity effect of statins). Statins may inhibit endogenous cholesterol production in the nerve cell body, creating a need to use exogenous cholesterol to maintain peripheral nerve membrane building. If exogenous LDL-C levels are also low as a result of statin therapy, this may impair nerve conduction (16,31). Consistent with this explanation, a longitudinal analysis of the Anglo-Danish-Dutch Study of Intensive Treatment of Diabetes in Primary Care (ADDITION) cohort reported a lower risk of DPN with higher levels of LDL-C (6). However, the slightly increased risk of DPN in new statin users in our study was driven by patients who initiated low-dose (<40 mg), not high-dose, statin therapy, which does not support a toxic effect. Also, observed median LDL-C levels after 1 year were 2.0 mmol/L (IQR 1.6–2.6) for new statin users, which is well within the recommended reference levels and likely enough for nerve cholesterol supply. Finally, intensive reduction of LDL-C levels with the newer proprotein convertase subtilisin/kexin type 9 inhibitors did not yield any adverse neuropathy symptoms (32).

Another DPN-mediating pathway may be the ability of statins to induce vasodilation by upregulation of endothelial nitric oxide synthase and, hence, increase the blood flow through nerve capillaries (10). At the time of type 2 diabetes diagnosis, nerve oxygen uptake from blood may already be impaired because of the accumulated effects of type 2 diabetes risk factors on nerve capillary function (7). Paradoxically, amplified nerve blood flow in a state of capillary dysfunction is expected to elicit oxidative stress, microvascular injury, and pain, potentially unmasking and worsening already evolving DPN (7). Meanwhile, the capillary function of peripheral nerves, theoretically, benefits from the improvement in blood viscosity induced by lowering triglyceride and cholesterol levels with statin therapy, just as intrinsic endothelial nitric oxide production in nerve microvessels is likely to adjust to optimize oxygenation over time (7). We speculate that such opposing effects of statin therapy, when initiated after the onset of type 2 diabetes, may contribute to a temporary worsening of DPN. However, despite potential mechanisms for statins to harm or support the nerve microenvironment, the overall data support no effect of statin therapy on DPN risk.

Biological sex appeared to modify the impact of statin therapy on hospital-diagnosed DPN risk. This effect measure modification was either not found (12) or not examined in previous studies (11,13). One possible explanation may be that females initiate statin therapy when they have more components of the metabolic syndrome. Statin initiation would then be a marker of a worse DPN risk profile compared with male statin initiators (4,33). Further research on this topic is needed.

Strengths of our study include the use of population-based registries in a tax-supported health care system with complete follow-up, which reduces the risk of referral and selection biases. Diagnosis codes with moderate to high PPVs were used to identify both the hospital-diagnosed DPN outcome and the diabetes population, the latter including patients with diabetes treated both in the primary care and the secondary care setting (19,26). Although we limited the inclusion of patients with type 1 diabetes by excluding those aged <30 years at diabetes diagnosis, our study population may still include some patients with late-onset type 1 diabetes, latent autoimmune diabetes of adulthood, monogenic diabetes, or pancreatogenic or drug-induced diabetes in addition to patients with “classical” type 2 diabetes.

Our study also has other limitations. First, DPN may be a difficult clinical diagnosis in routine care, depending on careful screening and multimodal clinical assessments (1). The slow progression of neuropathy symptoms, the absence of a disease-specific treatment, and incomplete coding may all have led to underdiagnosing DPN and, thus, a likely overrepresentation of DPN cases from the severe end of the spectrum in our study. We did not investigate the sensitivity of our DPN algorithm. However, in a bias analysis where we assumed a very low sensitivity (20%) for DPN, our risk estimates moved further toward the null, suggesting conservative bias and that our analyses were robust against incomplete records of DPN. Still, we cannot exclude potential detection bias if statin status was associated with the frequency of DPN examination. A second concern is that a high proportion (45%) of those who were statin nonusers at baseline initiated statin therapy during follow-up. This may have caused us to underestimate any statin effect, and we therefore supplemented our main analysis with an on-treatment analysis. This analysis yielded a 17% increased risk of DPN for new users, driven particularly by increased risk during the first year of follow-up. Third, as in any observational study, we cannot exclude residual confounding. We relied on hospital diagnosis codes for obesity, and incomplete adjustment for increased obesity among statin users could have led to overestimation of any adverse effect of statins on DPN risk (34). We also lacked detailed data on socioeconomic status, physical activity, and exact smoking habits. However, we were able to adjust for surrogate measures of smoking. Finally, although we used a new-user design and extensive confounder adjustment to counteract healthy adherer bias, uncontrolled healthy user effects may still have influenced our results. However, Danish statin users have been documented to have more comorbidities and unhealthier lifestyle habits than statin nonusers (35).

In conclusion, our study suggests that among patients with newly diagnosed type 2 diabetes, statin therapy is unlikely to increase or mitigate DPN risk. A small acute harmful effect cannot be excluded. This is largely outweighed by the substantial clinical effect of statins in cardiovascular disease prevention.

Funding. Research reported in this publication is part of the International Diabetic Neuropathy Consortium Research Programme, which is supported by a Novo Nordisk Foundation Challenge Programme grant (NNF14OC0011633).

Duality of Interest. B.C.C. consults for a Patient-Centered Outcomes Research Institute grant, DynaMed, and performs medical legal consultations, including consultations for the Vaccine Injury Compensation Program. S.T.K. reports grants outside the submitted work from AstraZeneca and personal fees from Boehringer Ingelheim, MSD, Novo Nordisk, Sanofi, and Mundipharma but with no conflicts relevant to the current study. The Department of Clinical Epidemiology, Aarhus University Hospital, receives funding for other studies from companies in the form of research grants to (and administered by) Aarhus University. None of these studies have any relation to the current study. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. F.P.K. was responsible for data management and performed statistical analyses in close collaboration with D.H.C. and R.W.T. F.P.K., D.H.C., and R.W.T. contributed to the design of the study, prepared the first draft, and contributed to the discussion of the results. B.C.C. provided expert epidemiological, statistical advice concerning the study design and data analysis; contributed to the discussion; and critically reviewed and edited the manuscript. J.K. sampled the study population, provided statistical advice, and critically reviewed and edited the manuscript. S.T.K., S.H.S., E.L.F., L.Ø., H.A., T.S.J., and H.T.S. contributed to the discussion and interpretation of the results and critically reviewed and edited the manuscript. All authors approved the final version for submission. F.P.K. and R.W.T. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Preliminary findings from this study were presented at the 12th Annual Meeting of the Nordic PharmacoEpidemiological Network (NorPEN), Aarhus, Denmark, 14–15 November 2019; at the 36th International Conference on Pharmacoepidemiology & Therapeutic Risk Management (ICPE), 16–17 September 2020; and at the 56th European Association for the Study of Diabetes (EASD) Annual Meeting, 21–25 September 2020.