Type 2 Diabetes mellitus, antidiabetics, and the risk of amyotrophic lateral sclerosis

Abstract

Background:

Research on the link between Type 2 Diabetes mellitus (T2DM) and amyotrophic lateral sclerosis (ALS) has produced mixed results. The potential role of antidiabetic medications in ALS etiology is also unclear. To contribute to these discussions, we aimed to examine the connections between T2DM, antidiabetic medications, and ALS using data from a large Israeli health fund.

Methods:

504 ALS cases diagnosed in 2002–2018 and 42,873 matched controls were considered in this population-based nested case-control study. T2DM was ascertained using diagnosis codes, laboratory test results, and medication use history, employing a 3-year lag from initial ALS diagnosis date to minimize chances for reverse causation. Multivariable-adjusted odds ratios (OR) were estimated for the association between T2DM, antidiabetic medications, and ALS.

Results:

T2DM overall was not linked with ALS (multivariable-adjusted odds ratio (OR) = 0.94, 95% confidence interval (CI): 0.72–1.23). However, T2DM with a history of insulin use showed a protective association with ALS (OR = 0.29; 95% CI = 0.09–0.92) compared to the non-T2DM group. A similar trend of protective associations with ALS was observed for T2DM with history of use of other antidiabetic medications, but none were statistically significant, and all associations were further attenuated after adjusting for insulin use.

Conclusions:

We observe a potential protective effect of T2DM-linked insulin use on risk of ALS. Although caution is necessary due to the limited number of ALS cases with insulin exposure, the observed protective association may suggest a biological pathway worth exploring for future therapeutic development.

Introduction

Growing evidence has suggested that energy metabolism is impaired in Amyotrophic lateral sclerosis (ALS) and further understanding of this biological aspect may be key for the development of therapeutic strategies.(1) Energy homeostasis is achieved when the cellular uptake of nutrients, such as glucose and lipids, is properly regulated by the glucose-insulin axis.(2) Diabetes is a common metabolic disorder characterized by elevated glucose levels resulting from impaired insulin secretion, insulin action, or both.(3) While glucose intolerance has been observed in many ALS patients,(4) several epidemiological studies found diabetes to be protective for ALS,(5–8) although some reported contrary (7, 9) or null (10, 11) findings.

Variations in the type of diabetes considered and in the antidiabetic medication regimens across the different study populations could potentially explain the inconsistent findings reported in the literature. Specifically, there is some data suggesting that the pathophysiology of type 1 (T1DM) and type 2 (T2DM) diabetes mellitus may differ with regard to ALS risk, with T1DM potentially associated with increased risk and T2DM with reduced risk (5). However, most previous studies did not clearly distinguish between T1DM and T2DM in their analyses, potentially leading to misclassification and obscuring true associations. In this study, we aimed to address this limitation by applying comprehensive criteria to identify T2DM and explicitly excluding individuals with T1DM. By doing so, we sought to more accurately assess the relationship between T2DM and ALS risk.

Associations with diabetes could result from effects of medications used for treatment. Some animal studies have suggested that antidiabetics may be beneficial for ALS,(12–14) but this has only been studied by few epidemiological studies, with inconsistent findings.(15–19) Moreover, most of these studies only explored diabetes medications as a group,(17, 19) and studies that examined the associations with specific antidiabetic medications were still unable to disentangle whether any observed associations stemmed from the medications itself or diabetes.(15, 19) Our study aimed to address this issue by separately analyzing the association between T2DM and ALS, in the presence and absence of specific antidiabetic medication use, using data from a large Israeli electronic medical database. This approach allowed us to explore whether the observed associations may be attributable to the use of antidiabetic medications, the underlying diabetes itself, or both.

Methods

Study Population

Data were collected from Maccabi Healthcare Services (MHS), the second largest integrated healthcare organization in Israel that provides universal health coverage to 2.6 million active members. MHS data include information on physician diagnosis from primary care, specialists’ clinics, and hospitalization; medication dispensing records; laboratory test results; medical procedures and imaging; and sociodemographic details. Complete data from MHS’s computerized electronic medical system were available starting in 1998. ALS cases diagnosed before December 31, 2018, were identified based on the International Classification of Diseases, Ninth Revision (ICD-9) code 335.20. A total of 697 ALS cases were identified, and the initial date of ALS diagnosis was defined as the index date. Each case was matched to 100 controls who shared the same birth year, were alive, and ALS-free at the index date. This initial matching was done with minimal criteria because the data were used for several analyses, and additional restrictions were implemented subsequently. We excluded any ALS case (n=193) or control (n=26,627) without four years of complete follow-up prior to the initial date of ALS diagnosis (the index date) to minimize the inclusion of prevalent ALS cases and to ensure that each participant had at least four years of follow-up to capture diabetes-related exposure data. Thus, no cases were included before 2002. We further excluded those with type 1 diabetes mellitus (T1DM, criteria described in the Diabetes Ascertainment section below) because the pathophysiology of T1DM and T2DM is different and previous literature showed that the two types had different associations with ALS.(5) Thus, the final analysis included 504 ALS cases and 42,873 controls. The study was approved by the MHS institutional review board and the Office of Human Research Administration at the Harvard T.H. Chan School of Public Health.

T2DM Ascertainment (Study variables and definitions)

To compare findings with previous literature, we first identified T2DM solely based on diagnosis records, using relevant International Classification of Diseases, 9th Revision (ICD-9) codes from groups 249.x and 250.x. However, for the main analyses, we employed a more comprehensive definition for ascertaining diabetes. Participants were defined as having T2DM if they met at least one of the following criteria:(1) HbA1c ≥ 7.25%; (2) at least two random glucose ≥ 200 mg/dL with a minimum gap of 30 days between tests; (3) at least two dispensing records for oral hypoglycemic agents (OHAs) at different dates in a three months period, with at least one HbA1c test result ≥ 6.5% or at least one fasting glucose test result ≥ 125 mg/dL; (4) at least two purchases of insulin in a three months period in patients who are not pregnant (patients who entered the registry based on dispensing of insulin alone and who did not have additional dispensing records for insulin for 6 consecutive months within the year following the initial dispensing date were not considered as having diabetes); (5) A diabetes related diagnosis given by a family physician / pediatrician / diabetologist/ endocrinologist / or ophthalmologist with one of the following: a. HbA1c ≥ 6.5% within six months before or after diagnosis; b. two fasting glucose tests ≥ 125 mg/dl, one six months before or after diagnosis and one at any point in time. In contrast, participants who fulfilled either of the following criteria were defined as having T1DM and were excluded from the analysis: (1) received diabetes diagnosis at ≤ 26 years of age; (2) purchased insulin ≥ five times and did not purchase other OHAs, or if other OHAs were purchased, first purchase occurred after at least five consecutive insulin purchases and after at least one year in the registry; (3) hadabnormal autoantibodieslevels (islet cell autoantibodies, glutamic acid decarboxylase autoantibody)(20) or c-peptide(21). Some participants were classified as having an undetermined type of diabetes due to insufficient data for accurate classification. In sensitivity analyses, we excluded these diabetes patients to confirm the consistency of the results.

Antidiabetics

We collected information on the use of specific antidiabetic medications based on Anatomical Therapeutic Chemical (ATC) codes. We defined four main groups of medication exposure (1) insulins (ATC group code A10A); (2) biguanide (A10BA), where specifically metformin is the only available biguanide on the market in most countries (22) and is also found in various combinations of blood glucose lowering drugs (A10BD, where individuals who used a medication belonging to the combined group were separately counted for each medication component in the combination); (3) sulfonylurea (A10BB); and (4) other diabetes medications (A10BF: Alpha glucosidase inhibitors, A10BG: Thiazolidinediones, A10BH: Dipeptidyl peptidase 4 (DPP-4) inhibitors, A10BJ: Glucagon-like peptide-1 (GLP-1) analogues, A10BK: Sodium-glucose co-transporter 2 (SGLT2) inhibitors, A10BX: Other blood glucose lowering drugs).

Covariates

Age was inherently adjusted for in the matching process, and the following covariates were included in the fully adjusted model: sex, smoking (ever, never, and unknown), socioeconomic status (SES, on a 1–10 scale as previously described),(23) residential district (Jerusalem, Center, North, South, Sharon), and population subgroup (Israeli Arabs, Jewish Orthodox, and Secular).

Statistical analysis

Conditional logistic regression was used to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) for the association between T2DM and ALS. In main analyses, those with diabetes of unknown type were considered T2DM, but in sensitivity analyses we excluded these participants. Stratified analyses were conducted to investigate whether the associations between T2DM and ALS differed by sex. We further examined the associations between specific T2DM-related antidiabetics (i.e., insulin and OHAs—metformin, sulfonylurea, and other oral diabetic medications) and ALS. In these analyses, participants were categorized as (1) no T2DM, (2) T2DM without any history of antidiabetic medication use, (3) T2DM plus a history of a specific antidiabetic medication use (ever/never), and (4) T2DM with a history of any medication use, but not of the specific medication group considered in the category above. Separate models were run for each medication group, keeping the no T2DM and T2DM without any history of antidiabetic medication use the same across all models. For all the aforementioned analyses, sex, smoking status, SES, district, and population subgroup were included in the fully adjusted models. For analyses investigating the associations between OHAs and ALS, we also explored additional adjustment for insulin use because we saw a strong inverse association for insulin in all analyses.

Secondary analyses

According to clinical guidelines, insulin has been suggested as a third-to-fourth-line therapy and is typically initiated when glycemic goal is not achieved despite the use of combination therapy of two to three OHAs at their optimal doses for 3–6 months.(24) In order to tease out whether an association between insulin and ALS risk is related to polytherapy use, we further investigated a potential polytherapy effect, we categorized participants into (1) diabetes without antidiabetic medication use (2) diabetes with monotherapy but no insulin use (3) diabetes with polytherapy but no insulin use (4) diabetes with a history of insulin use, either alone or in polytherapy with other medications. In addition, because insulin is often initiated at an advanced diabetes stage while OHAs are typically preferred for first-line management of hyperglycemia,(24) we separately explored the association between diabetes duration and risk of ALS. Diabetes onset was considered as the first date of diabetes diagnosis or antidiabetic medication dispensing in the MHS data. Those whose diabetes onset was within 6 months of Jan 1, 1998, were defined as having an unknown duration because we could not know whether they had diabetes prior to the start of the electronic data. All analyses were performed in R version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Population Characteristics

Participant characteristics are shown in Table 1. The mean age at diagnosis for the 504 inluded ALS patients was 60.8 (SD=14.9) years and 42.5% were female. A higher percentage of unknown smoking status was observed for the ALS case, likely because routine data collection on smoking status started later in MHS and many of the ALS cases died before data on smoking were systematically collected. Among the 42,873 controls, 4,877 (11.4%) participants met the T2DM criteria in the MHS registry (online supplemental eTable 1). Participants without diabetes were generally younger.

Table 1.

Baseline Characteristics of ALS Cases and Matched Controls in the Maccabi Healthcare Services Database (2002-2018)

	ALS (n=504)	No ALS (n=42,873)
Age, mean (SD)	60.8 (14.9)	61.3 (14.6)
Female, N (%)	214 (42.5)	22,495 (52.5)
Smoking, N (%)
Ever	67 (13.3)	4,852 (11.3)
Never	344 (68.3)	24,035 (56.1)
Missing	93 (18.5)	13,986 (32.6)
SES, N (%)
<=4	73 (14.5)	7,904 (18.4)
5-6	197 (39.1)	17,910 (41.8)
7-8	143 (28.4)	11,383 (26.6)
9-10	66 (13.1)	5,233 (12.2)
Missing	22 (4.4)	443 (1.0)
Population N (%)
Arab	60 (11.9)	6,156 (14.4)
Orthodox	29 (5.8)	2,435 (5.7)
Secular	415 (82.3)	34,282 (80.0)
District, N (%)
Jerusalem & Shfela	115 (22.8)	10,329 (24.1)
Center	121 (24.0)	9,305 (21.7)
North	81 (16.1)	8,329 (19.4)
South	60 (11.9)	6,578 (15.3)
Sharon	108 (21.4)	8,118 (18.9)
Missing	19 (3.8)	214 (0.5)

ALS, amyotrophic lateral sclerosis; SES, socioeconomic status.

Smoking status was categorized as: ever (current or former), never, and unknown.

T2DM and ALS

We did not find evidence of an association between T2DM and the risk of ALS, irrespective of whether T2DM was defined using the ICD-9 diagnosis codes only (OR, 1.05; 95% CI, 0.82–1.35) or based on the MHS registry (OR, 0.94; 95% CI, 0.72– 1.23) (Table 2). When stratified by sex, the associations trended in opposite directions with reduced odds seen among females and increased odds in males (multivariable adjusted OR, 0.67; 95% CI, 0.42–1.07 and OR, 1.17; 95% CI, 0.83–1.63, respectively). The p-value for the interaction term between sex and T2DM was 0.02.

Table 2.

Odds ratios and 95% confidence intervals for type 2 diabetes mellitus (T2DM), ascertained using diagnosis codes only or based on diagnosis codes, lab results, and medication dispensing data.

	Case	Control	OR (95% CI)
			Age, sex adjusteda	MVb Adjusted
Diabetes based on diagnosis codes only
Overall
No DM	418	36,545	Ref	Ref
DM	86	6,328	1.23 (0.96-1.57)	1.05 (0.82-1.35)
Female
No DM	184	19,270	Ref	Ref
DM	30	3,225	0.98 (0.66-1.48)	0.85 (0.56-1.28)
Male
No DM	234	17,275	Ref	Ref
DM	56	3,103	1.44 (1.06-1.96)	1.24 (0.90-1.69)
Diabetes based on diagnosis codes, lab results, and medication dispensing data c
Overall
No DM	435	37,315	Ref	Ref
DM	69	5,558	1.09 (0.83-1.42)	0.94 (0.72-1.23)
Female
No DM	192	19,646	Ref	Ref
DM	22	2,849	0.78 (0.50-1.24)	0.67 (0.42-1.07)
Male
No DM	243	17,669	Ref	Ref
DM	47	2,709	1.35 (0.97-1.88)	1.17 (0.83-1.63)

^aAge was inherently adjusted via matching. The “Age + Sex adjusted” model includes additional adjustment for sex.

^bMV: Multivariable-adjusted (adjusted for age via matching, with further adjustment for sex, smoking, SES, district of residence, and population subgroup.)

^cT2DM was defined using comprehensive criteria including: (1) HbA1c ≥ 7.25%; (2) two random glucose readings ≥ 200 mg/dL at least 30 days apart; (3) ≥2 dispensings of oral hypoglycemic agents within 3 months, plus either HbA1c ≥ 6.5% or fasting glucose ≥ 125 mg/dL; (4) ≥2 insulin dispensings within 3 months (excluding pregnancy), with sustained use; or (5) a physician diagnosis with confirmatory lab results (e.g., HbA1c ≥ 6.5% or two fasting glucose tests ≥ 125 mg/dL). Individuals meeting criteria for T1DM (early onset, exclusive or early insulin use, or abnormal autoantibodies/c-peptide) were excluded.

Antidiabetics and ALS

Compared to those without T2DM, the OR for T2DM patients without a history of antidiabetic use was slightly elevated (OR, 1.27; 95% CI, 0.74–2.18) while among those taking any antidiabetic medications it was slightly lower (OR, 0.88; 95% CI, 0.66–1.18) (Table 3). When T2DM antidiabetic users were split into those who used insulin and those who used other antidiabetics and compared to those without T2DM, those who used insulin had a significantly lower odds of ALS (OR, 0.29; 95% CI, 0.09–0.92), whereas no association was observed among T2DM patients who used other antidiabetics (OR, 1.00; 95% CI, 0.74–1.35). The ORs were more attenuated when considering T2DM patients who used other OHAs, and were further attenuated towards the null with additional adjustment for insulin co-use. These results were largely similar by sex, although the inverse association with insulin was more pronounced among males (online supplemental eTable 2 and eTable 3). In addition, the ORs for the different medication groups tended to be larger for males compared to females, particularly so for the “other medications” category. In sensitivity analyses that excluded those without smoking data, the results were essentially unchanged (online supplemental eTable 4). In sensitivity analyses that excluded those with unknown diabetes type, results were also similar, although the inverse association with insulin use was somewhat attenuated (OR, 0.42; 95% CI, 0.14–1.33) (online supplemental eTable 5).

Table 3.

Association between type 2 diabetes mellitus (T2DM) and ALS by history of antidiabetic medication use.

T2DM & Medication group	Case	Control	OR (95% CI)
			Age, sex adjusted*	MV** Adjusted	MV+insulin use
No T2DM	435	37,315	Ref	Ref	Ref
T2DM, no diabetes medications	14	830	1.44 (0.84-2.48)	1.27 (0.74-2.18)	-
Any antidiabetic
T2DM, any diabetes medications	55	4,728	1.02 (0.76-1.37)	0.88 (0.66-1.18)	-
Insulin
T2DM with medications, other than insulin	52	3,975	1.15 (0.85-1.55)	1.00 (0.74-1.35)	-
T2DM with insulin	3	753	0.35 (0.11-1.09)	0.29 (0.09-0.92)	-
Metformin
T2DM with medications other than metformin	7	623	0.95 (0.45-2.03)	0.83 (0.39-1.78)	1.03 (0.48-2.23)a
T2DM with metformin	48	4,105	1.03 (0.76-1.41)	0.89 (0.65-1.22)	0.99 (0.72-1.36)a
Sulfonylurea
T2DM with medications other than sulfonylurea	29	2,264	1.14 (0.78-1.68)	0.98 (0.66-1.44)	1.06 (0.72-1.58)b
T2DM with sulfonylurea	26	2,464	0.91 (0.61-1.37)	0.80 (0.53-1.20)	0.93 (0.61-1.41)b
Other diabetic medications
T2DM with either insulin, metformin or sulfonylureas	37	3,182	1.02 (0.73-1.45)	0.89 (0.63-1.27)	0.97 (0.69-1.38)c
T2DM with other antidiabetic meds	18	1,546	1.02 (0.63-1.65)	0.86 (0.53-1.40)	1.05 (0.64-1.72)c

^*Age was inherently adjusted via matching. The “Age + Sex adjusted” model includes additional adjustment for sex.

^**Adjusted for age via matching, with further adjustment for sex, smoking, SES, district of residence, and population subgroup

^aOR (95% CI) for insulin from this model: 0.30 (0.09-0.96)

^bOR (95% CI) for insulin from this model: 0.30 (0.09-0.98)

^cOR (95% CI) for insulin from this model: 0.29 (0.09-0.95)

We additionally explored whether the inverse association with insulin was related to polytherapy use. Compared to those without T2DM, no significant association was seen for T2DM patients with mono- or polytherapy that excluded insulin, whereas T2DM patients with a history of insulin use still had a significant inverse association (OR, 0.30; 95% CI, 0.09–0.93) (Table 4).

Table 4.

Association between type 2 diabetes mellitus (T2DM) and ALS by history of insulin use, considering co-exposure to other antidiabetics in mono or polytherapy.

	Case	Control	OR (95% CI)
			MV Adjusted*
No DM	435	37,315	Ref
T2DM, no diabetes medications	14	830	1.27 (0.74-2.18)
T2DM with monotherapy, but no insulin	28	1,960	1.10 (0.74-1.64)
T2DM with polytherapy, but no insulin	24	2,015	0.90 (0.59-1.37)
T2DM with insulin	3	753	0.30 (0.09-0.93)

^*Adjusted for age via matching, with further adjustment for sex, smoking, SES, district of residence, and population subgroup

We further explored whether the inverse association for insulin was related to longer T2DM duration. T2DM patients without antidiabetic use had the shortest duration (mean [SD], 6.4 [3.0] years), followed by patients with OHAs use (mean [SD), 7.9 [4.9] years), and patients with insulin use had the longest duration (mean [SD], 11.1 [5.7] years). There was no obvious trend in the ORs of ALS by T2DM duration, which remained the case when insulin use was added to the model. In this model the OR for insulin use was still protective (OR, 0.33; 95% CI, 0.10–1.08) (Table 5).

Table 5.

Association between type 2 diabetes mellitus (T2DM) duration and ALS.

	Case	Control	OR (95% CI)
			MV Adjusted*	MV+insulin use**
No T2DM	435	37,315	Ref	Ref
T2DM duration quartile 1	13	989	1.02 (0.58-1.78)	1.07 (0.61-1.88)
T2DM duration quartile 2	14	988	1.03 (0.60-1.78)	1.06 (0.61-1.83)
T2DM duration quartile 3	16	985	1.24 (0.75-2.08)	1.29 (0.77-2.15)
T2DM duration quartile 4	13	989	1.01 (0.57-1.80)	1.13 (0.63-2.01)
Unknown T2DM duration	13	1607	0.61 (0.35-1.07)	0.76 (0.42-1.35)

^*Adjusted for age via matching, with further adjustment for sex, smoking, SES, district of residence, and population subgroup

^**OR (95% CI) for insulin from this model: 0.33 (0.10-1.08)

Discussion

Using data from the large Maccabi healthcare database, we did not find strong evidence for an association between T2DM and ALS. This was true regardless of whether T2DM was defined based on diagnosis codes only, or a more comprehensive ascertainment criteria that additionally used data on laboratory tests results and medication dispensing records. Nonetheless, the OR for ALS was somewhat elevated for T2DM patients who were unmedicated, and slightly protective among those taking any antidiabetic medications. Further analyses indicated that the protective association was largely attributed to insulin, and that it was not explained by exposure to polytherapy treatment regimen or a longer duration since T2DM onset. While this was the only significant association among several subgroup analyses, the finding suggests a potentially interesting lead to follow-up on in the search for prevention opportunities.

Current literature on the association between diabetes and ALS is inconclusive. Several studies from European populations reported inverse associations between diabetes and ALS,(5–8) whereas positive associations were observed in Asian populations,(7, 9) and one study from the Netherlands and one from Taiwan found null results.(10, 11) Studies that further examined the associations between T1DM versus T2DM mostly found that only T2DM was associated with a lower risk of ALS, whereas T1DM had a positive association.(5, 6, 25) These epidemiological findings were further supported by Mendelian Randomization (MR) studies, which provided evidence for a potential causal relationship between T2DM and ALS. Zeng et al. (7) reported a protective association between genetically predicted T2DM and ALS in European populations, although a suggestive positive association was observed in East Asians. A subsequent MR study by Liu et al. (26) confirmed an inverse association in both populations. These MR findings strengthen the plausibility of a protective role for T2DM in ALS etiology by reducing confounding and reverse causation concerns inherent to observational studies.

Epidemiological studies on the associations between antidiabetics and the risk of ALS have remained limited and inconsistent. One study from Germany(17) and one from the Netherlands(18) found no association between overall antidiabetic medication use and ALS. Two studies that examined the associations between specific antidiabetic medications with ALS reported that insulin,(15) sulfonylureas,(15) and metformin(15, 19) had inverse associations. Findings from the current study showed a consistent protective association between insulin and ALS. Although insulin use may be related to multiple OHA use and longer T2DM duration,(24) the null findings in our polytherapy and diabetes duration analyses suggested that the inverse association we observed may be with insulin per se. Findings from the current study raise the possibility that the inconsistencies in prior studies may be related to the differing prescribing practices of antidiabetics, including the timing of insulin initiation across populations and over time. Further, a protective effect of insulin may be limited to patients with T2DM, and inconsistencies concerning the exclusion of T1DM patients may have obscured any protective effects linked with insulin use. Our use of comprehensive ascertainment criteria for diabetes was an advantage in this regard, as solely relying on diagnosis codes for deriving diabetes type can lead to substantial classification error.(27, 28)

Araki et al. reported that TDP-43 (TAR DNA-binding protein 43 kDa), the nuclear loss of which is one of the histopathological hallmarks of ALS, was a novel regulator for insulin secretion. Their findings indicated that the nuclear loss of TDP-43 may be responsible for both selective motor neuron death and impaired/reduced insulin secretion in the early phase of ALS.(29) Bilic et al. showed that ALS patients had significantly lower levels of insulin both in the serum and cerebrospinal fluid (CSF) compared with normal controls; they also found that the CSF level of insulin-like growth factor-1 (IGF-1), a hormone that has a similar molecular structure and hypoglycemia effects as insulin,(30) was lower in ALS patients.(31) IGF-1 has been considered an important neurotrophin, contributing to motor neuron maintenance and survival.(32, 33) Although some animal studies found that IGF-1 improved survival and delayed disease progression in animal ALS models,(32, 33) studies in humans remain inconclusive. Among the three human RCTs that examined the effect of subcutaneous IGF-1 on ALS progression, one study showed that patients in the high-dose group had slower disease progression and less decline in quality of life,(34) whereas the other two studies found no beneficial effects.(35, 36) Subcutaneous delivery, dosage of IGF-1, and treatment duration may be the reasons for the null effects in the aforementioned studies.(37) In addition, one recent study found that systemic insulin injection improved survival in an animal ALS model.(38) A recent drug-target MR study only reported a protective association for SGLT2 inhibition but did not find a significant causal association between genetically proxied insulin targets and ALS risk (39). The null finding for insulin in this MR study may reflect limitations in the availability and specificity of genetic instruments for capturing insulin use or activity. Taken together, current evidence highlights the need for further investigation to determine whether insulin exerts a true protective effect against ALS.

Strengths of the current study include the use of a nationally representative cohort and excellent ascertainment of T2DM and antidiabetic medications by medication dispensing records, clinical diagnoses, and lab results. This is particularly critical for excluding T1DM patients, since these patients would be treated with insulin, but have also been found to have a higher risk of ALS,(5, 6) which could obscure a protective effect of insulin. We also had long follow-ups on study participants to assess the exposures and good information on demographic factors that can possibly act as confounders. There are also several limitations to the current study. First, we only focused on T2DM but not related comorbidities, therefore, etiological involvement of other T2DM-related comorbidities in ALS etiology is possible. Second, our data started in 1998, therefore we were unable to assess early life exposure. Third, we were unable to investigate specific ALS subtypes. Fourth, despite the use of comprehensive criteria for ascertaining diabetes type, we were not able to classify the diabetes subtype for all patients, and misclassification of some diabetes patients as T2DM is possible. However, the results were consistent when excluding patients whose diabetes status could not be classified, and misclassification of some T1DM as T2DM would have likely attenuated the OR observed for insulin use, thus strengthening our finding of a protective association with insulin.(5, 6, 25) Fifth, we did not have consistent HbA1c data across the full cohort to adjust for long-term blood glucose control, which may be associated with both the intensity of diabetes treatment and overall mortality thus resulting in an apparent protective association for medications associated with worse glycemic control, like insulin. However, polytherapy use would also be a proxy for diabetes severity and glycemic burden, and the findings among those using polytherapy were null, suggesting that worse glycemic control was not accounting for our findings with insulin use. Sixth, we lacked consistent data on body mass index (BMI), and higher BMI is a known side effect of insulin use that has been associated with a reduced risk of ALS. Future studies with detailed and longitudinal anthropometric data will be important to understand whether an association with insulin may result from effects on BMI. Lastly, the number of ALS cases that used insulin was small (n=3). However, if in fact insulin is protective for ALS, then we would expect fewer ALS cases who used insulin. Given the OR for ALS among insulin users of about 0.3, had there been no association between insulin and ALS we would have expected about 10 ALS cases who used insulin. Relatedly, we did not correct for multiple comparisons as this was a first look at these questions meant to identify directions for further research. Follow-up in even larger cohorts so that the number of insulin users is larger is needed.

In conclusion, results from the current study suggested a possible protective association between T2DM-linked insulin use and ALS. We would expect these results to be generalizable to other populations, and could contribute to explaining prior discrepancies in findings for diabetes and ALS. Future research is needed to investigate the mechanisms and effect of insulin on ALS risk and survival as this could suggest potential therapeutic targets.

Data availability statement:

Data used in this work are protected by privacy laws. Access is contingent upon obtaining the required IRB and administrative approvals by Maccabi Healthcare Services. More information could be obtained by contacting Dr. Rotem at ran.rotem@mail.harvard.edu