Type 2 diabetes, obesity, and risk of amyotrophic lateral sclerosis: A population‐based cohort study

Abstract Background Type 2 diabetes and obesity may be inversely associated with amyotrophic lateral sclerosis (ALS), but the evidence is controversial. Methods Using Danish, nationwide registries (1980‐2016), we identified patients with a diagnosis of type 2 diabetes (N = 295,653) and patients with a diagnosis of obesity (N = 312,108). Patients were matched (1:3) to persons from the general population on birth year and sex. We computed incidence rates and Cox regression derived hazard ratios (HRs) of a diagnosis of ALS. In multivariable analyses, HRs were controlled for sex, birth year, calendar year, and comorbidities. Results We observed 168 incident cases of ALS (0.7 [95% confidence interval (CI): 0.6–0.8] per 10,000 person‐years) among patients with type 2 diabetes and 859 incident cases of ALS (0.9 [95% CI: 0.9–1.0] per 10,000 person‐years) among matched comparators. The adjusted HR was 0.87 (95% CI: 0.72–1.04). The association was present among men (adjusted HR: 0.78 [95% CI: 0.62–0.99]) but not women (adjusted HR: 1.03 [95% CI: 0.78–1.37]), and among those aged ≥60 years (adjusted HR: 0.75 [95% CI: 0.59–0.96]) but not younger. We observed 111 ALS events (0.4 [95% CI: 0.4–0.5] per 10,000 person‐years) among obesity patients and 431 ALS events (0.5 [95% CI: 0.5–0.6] per 10,000 person‐years) among comparators. The adjusted HR was 0.88 (95% CI: 0.70–1.11). Conclusions Diagnoses of type 2 diabetes and obesity were associated with a reduced rate of ALS compared with general population comparators, particularly among men and patients aged 60 years or above. However, absolute rate differences were small.


INTRODUCTION
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease with insidious onset and characterized by progressive loss of motor neurons in the brain and spinal cord, which usually culminates in death (van Es et al., 2017). The incidence is higher in men than in women, and the prognosis is poor. The median survival from symptom onset has been reported to be 32 months (del Aguila et al., 2003) and respiratory paralysis is a common cause of death (Brown & Al-Chalabi, 2017).
For example, in a Danish case-control study,  found a large protective effect of diabetes (odds ratio: 0.61 [95% confidence interval (CI): 0.46−0.80]). In a Swedish case-control study, Mariosa et al. (2015) found a similar effect size (odds ratio: 0.79 [95% CI: 0.68−0.91]). Three cohort studies, two from Taiwan (Sun et al., 2015;Tsai et al., 2019) and one from Italy (D'Ovidio et al., 2018), showed divergent results (hazard ratio point estimates ranging from 0.30 to 1.35). A systematic review concluded that the evidence of an association between diabetes and ALS risk is limited, calling for further evidence from cohort studies (Lekoubou et al., 2014).
We examined the association between type 2 diabetes and ALS and that between obesity and ALS using Danish registry data. We hypothesized that hospital-based diagnoses of type 2 diabetes and obesity were inversely associated with incident ALS.

Design and setting
We conducted a population-based matched cohort study in Denmark based on nationwide administrative registry data between January 1, 1980 and December 31, 2016. Denmark has a tax-supported health care system, ensuring equal access to general practice and hospitalbased care (Schmidt et al., 2019). A unique personal identification number, assigned to all residents at birth or upon immigration, permits unambiguous, individual-level linkage across data sources (Schmidt et al., 2014).

Type 2 diabetes and obesity cohorts
We used the Danish National Patient Registry (DNPR) (Schmidt et al., 2015) to identify a cohort of patients with a first-time hospital-based diagnosis of type 2 diabetes mellitus and another cohort of patients with a first-time hospital-based diagnosis of obesity. For both cohorts, we searched for both primary and secondary and inpatient and outpatient clinic diagnoses between January 1, 1980 and December 31, 2016. We excluded patients with a prior diagnosis of ALS or other motor neuron diseases (122 diabetes patients and 71 obesity patients).
The DNPR is an ongoing administrative registry that contains records of all admissions and discharges from nonpsychiatric hospitals since 1977 and of all outpatient and emergency clinic visits since 1995 (Schmidt et al., 2015). Each hospital discharge or outpatient clinic visit is recorded with one primary diagnosis and up to 19 secondary diagnoses, coded according to the International Classification of Diseases 8th revision (ICD-8) from 1977 to 1993 and the 10th revision (ICD-10) thereafter.
Although we searched for hospital-based diagnostic codes specific to type 2 diabetes, it is possible that some type 1 diabetes patients were erroneously coded with type 2 diabetes. To reduce this possible source of exposure misclassification, we constructed a subcohort consisting of patients diagnosed with type 2 diabetes who previously had redeemed a prescription for a noninsulin glucose-lowering drug (e.g., metformin).
Metformin is the recommended first-line treatment for type 2 diabetes in Denmark (Christensen et al., 2016), and thus, it is unlikely that this subcohort contained patients with type 1 diabetes. As prescription data were available only from 2004 and onward, this subcohort was restricted to patients first diagnosed from 2005 and onward. We identified redeemed prescriptions from the Danish National Health Service Prescription Data (Johannesdottir et al., 2012), which contains data on all drug prescriptions redeemed in community pharmacies since January 1, 2004. Available data for each prescription include the date of redemption, the Anatomical Therapeutic Chemical Classification System code, type, and quantity.
In a validation study against general practice records, the positive predictive value of a diabetes diagnosis in the DNPR was found to be > 90% (Carstensen et al., 2011). Similarly, when measured against computerized height and weight measurements from hospital contacts, the positive predictive value of an obesity diagnosis in the DNPR was found to be 88% (Gribsholt et al., 2019).

General population comparison cohorts
To compare the risk of ALS after a type 2 diabetes or obesity diagnosis with that in the general population, we used the Civil Registration System (Schmidt et al., 2014)

Amyotrophic lateral sclerosis
The primary outcome was a primary or secondary hospital-based inpatient or outpatient clinic discharge diagnosis of ALS, as recorded in the DNPR with ICD-8 code 348 or ICD-10 code G12.2 (ALS and other motor neuron diseases). A validation study found the positive predictive value of this definition to be 78% (95% CI: 71−84) when measured against medical records and applying the El Escorial criteria . However, when also including clinically suspect ALS cases, the positive predictive value was 93% (95% CI: 88−96). Although the positive predictive value of this definition is reasonably high, in a supplementary analysis we also considered a stricter definition of ALS (ICD-8 code 348.09, ICD-10 code G12.2G) that excluded related forms of motor neuron disease.

Covariates
Smoking is associated with an increased risk of diabetes (Haire-Joshu et al., 1999) and may also increase ALS risk (de Jong et al., 2012). We have previously shown that statins (primary and secondary therapeutic for patients with arterial cardiovascular disease) may increase ALS risk, especially in women (Skajaa et al., 2021). Thus, based on hospital-based discharge diagnoses registered before the index date in the DNPR, we obtained information on chronic obstructive pulmonary disease (as an indicator of sustained smoking), myocardial infarction, stroke, hypercholesteremia, hypertension, atrial fibrillation, heart failure, cancer, and chronic kidney disease.

Statistical analyses
As ALS has a preclinical diagnostic phase, which often leads to a diagnostic delay (Eisen et al., 2014), we excluded the first year following the type 2 diabetes or obesity diagnosis to mitigate the potential identification of prevalent ALS diagnoses and to reduce the potential risk of surveillance bias associated with contact with the healthcare system. Thus, follow-up started 1 year from the date of diagnosis until the first occurrence of a hospital-based ALS diagnosis, death, emigration, or administrative study end (December 31, 2016). If a person from a comparison cohort was diagnosed with diabetes or obesity during follow-up, he or she was transferred to the respective patient cohort and censored from the comparison cohort. For the type 2 diabetes analysis, we performed all analyses for both the overall cohort and the subcohort (2005−2016) of patients on glucose-lowering drugs. We calculated the median follow-up time, the number of ALS events, and incidence rates per 10,000 person-years. We plotted the cumulative incidence of ALS, treating death as a competing risk (Andersen et al., 2012). We then used stratified Cox proportional hazards regression to calculate unadjusted and adjusted hazard ratios (HRs) of ALS, comparing the type 2 diabetes cohort with general population comparators and the obesity cohort with general population comparators. Time since the index date was the underlying time-scale. In both unadjusted and adjusted models, the matching factors (birth year, sex, and calendar year) were controlled for by design, and not included in the models. The adjusted models additionally included the covariates described above.
When stratifying by sex, age, and calendar period, stratified Cox analyses were used; otherwise, ordinary Cox analyses were used, in which the matching was dissolved and the matching factors instead included in the model.

Sensitivity analyses
We performed a number of sensitivity analyses to examine the robustness of our results. First, we repeated the analyses using a stricter definition of ALS, that is, using codes specifically relating to ALS and not to other motor neuron diseases. Second, for patients diagnosed with type 2 diabetes or obesity during 2005−2016 (the period prescription data were available), we additionally adjusted for the use of statins and antihypertensives.
All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC). The study was approved by the Danish Data Protection Agency (record no. 1-16-02-1-08). Table S1 lists all codes used in the study.

RESULTS
The analyses

Type 2 diabetes versus general population
In the overall cohort of patients with type 2 diabetes, we observed 168 and not in women (
Although estimates were fairly imprecise, we observed a similar pattern of effect measure modification for obesity as for type 2 diabetes: Adjusted HRs were lower in men than women, in those aged ≥60 years than those younger, and in patients diagnosed during 1994-2016.
However, the HR appeared to increase with an increasing number of comorbidities, in contrast to that observed for type 2 diabetes.

DISCUSSION
In this large population-based cohort study with more than 30 years of observation time, an inpatient or outpatient hospital-based diagnosis of type 2 diabetes was associated with a reduced rate of ALS compared with matched general population comparators, specifically among patients aged 60 years or above and more clearly in men than women. A hospital-based diagnosis of obesity was also associated with a marginally reduced rate of ALS, but the effect estimate was less precise. Despite the observed protective effects, the absolute rate differences were small.
Some potential limitations must be discussed. First, ALS has an insidious onset, which may have led to the inclusion of diabetes and obesity patients with prevalent ALS. However, we mitigated this potential risk, as well as the potential risk of surveillance bias arising from contact with the healthcare system associated with the diagnostic workup of diabetes or obesity, by excluding from the analysis the first year following diagnosis. Second, we cannot rule out misclassification of both the exposure (i.e., type 2 diabetes or obesity) and outcome. While the completeness of the obesity diagnosis is low in the DNPR, the positive predictive value of both diabetes and obesity diagnoses is high (88−90%) (Carstensen et al., 2011;Gribsholt et al., 2019). Regardless, any potential exposure misclassification would likely bias the associations toward unity, that is, masking a true protective effect.
The analysis of the subcohort of type 2 diabetes patients on noninsulin glucose-lowering drugs, undertaken to reduce misclassification with type 1 diabetes, showed a lower hazard, further demonstrating that the overall results may have been conservative. On the other hand, the association decreased with increasing calendar period, and patients comprising the subcohort were diagnosed in more recent years; thus, the more pronounced findings for the subcohort could also be explained by calendar time. Our patient inclusion was restricted to those with an inpatient or outpatient hospital-based diagnosis, thereby not capturing patients with mild disease. However, we probably identified some of these patients in the subcohort of type 2 diabetes patients on glucose-lowering drugs, because prescription data are available for the entire Danish population. Third, as our study was observational, we cannot rule out residual or unknown confounding. Of note, we proba-F I G U R E 2 Events, incidence rates, and adjusted hazard ratios, comparing type 2 diabetes patients with a previously redeemed glucose-lowering drug with matched general population comparators, overall and in subgroups according to sex, age groups, calendar period, and number of comorbidities. T2D: type-2 diabetes; GP: general population; PY: person-years; HR: hazard ratio. Adjusted HRs were for controlled for matching factors (sex, age, calendar year) and additionally adjusted for chronic obstructive pulmonary disease, myocardial infarction, stroke, hypercholesteremia, hypertension, atrial fibrillation, heart failure, cancer, chronic kidney disease. Comorbidities include chronic obstructive pulmonary disease, myocardial infarction, stroke, hypercholesteremia, hypertension, atrial fibrillation, heart failure, cancer, chronic kidney disease, and obesity.
F I G U R E 3 Cumulative incidence of amyotrophic lateral sclerosis among patients with type 2 diabetes, patients with type 2 diabetes with a noninsulin glucose-lowering drug (subcohort), patients with obesity, and matched general population comparators.
bly only partly adjusted for the confounding effect of smoking, using a hospital-based diagnosis of chronic obstructive pulmonary disease as a proxy for sustained smoking. We also lacked data on physical activity. However, the small differences between unadjusted and adjusted HRs leave little indication that residual confounding could explain the results.
Our findings agree with most previous literature indicating that cardiometabolic conditions, including type 2 diabetes and obesity, are associated with a reduced risk of ALS risk (D'Ovidio et al., 2018;Mariosa et al., 2015;O'Reilly et al., 2013;Tsai et al., 2019). The effect sizes observed in this study regarding both associations are at par with those reported earlier. However, unlike previous reports, we found an indication of a possible sex difference, as the protective association of type 2 diabetes in the overall cohort was present in men and not in women. We are unaware of any explanation for this finding. Nevertheless, we have previously found a large sex-differential effect when examining the effect of statins on ALS risk, as the observed adverse effect of statins was present only in women (Skajaa et al., 2021). In alignment with existing knowledge Mariosa et al., 2015;Tsai et al., 2019), we observed clear effect modification by age, as any type 2 diabetes benefit was present only in those aged 60 years or older.  hypothesized that the null or harmful association found for younger ages could be explained by differences between type 1 and 2 diabetes, that is, a type 1 diabetes diagnosis is most often given at younger ages. As noted above, we cannot rule out misclassification between type 1 and 2 diabetes despite only including diagnostic codes specific for type 2 diabetes. However, although estimates were much F I G U R E 4 Events, incidence rates, and adjusted hazard ratios, comparing obesity patients with matched general population comparators in subgroups according to sex, age groups, calendar period, and number of comorbidities. GP: general population; PY: person-years; HR: hazard ratio. Adjusted HRs were for controlled for matching factors (sex, age, calendar year) and additionally adjusted for chronic obstructive pulmonary disease, myocardial infarction, stroke, hypercholesteremia, hypertension, atrial fibrillation, heart failure, cancer, chronic kidney disease. Comorbidities include chronic obstructive pulmonary disease, myocardial infarction, stroke, hypercholesteremia, hypertension, atrial fibrillation, heart failure, cancer, chronic kidney disease, and obesity. less precise, the same age pattern was observed for obesity, which may warrant another explanation.
The mechanisms explaining the theoretical benefits of cardiometabolic conditions, including type 2 diabetes and obesity, are not well-understood. ALS appears to be a disease not restricted to the central nervous system but may have more widespread effects, including effects on energy metabolism (Dupuis et al., 2011). In most ALS patients, energy uptake is reduced while energy expenditure is increased, leading to reduced fat depots (Vandoorne et al., 2018). This may explain the protective effects observed for obesity; however, assuming causality, the direction of the cause-effect association is convoluted as the ALS onset is insidious, and many patients may simply be pre-symptomatic for ALS, and thus undiagnosed, at the time of obesity diagnosis. On the other hand, such an effect could also be explained by elevated lipid and cholesterol levels (Dupuis et al., 2008;Sutedja et al., 2011). Regarding diabetes, most evidence suggests that a potential protective effect on ALS risk is restricted to type 2 diabetes, a condition associated with insulin resistance Mariosa et al., 2015). It is possible that metformin treatment, the first-line treatment of type 2 diabetes in Denmark (Christensen et al., 2016), could exert some benefit. While animal studies have yielded mixed results (Kaneb et al., 2011;Sun et al., 2013), a Swedish case-control study suggested that patients using antidiabetic drugs had a 35% to a 10% lower risk of ALS compared with matched nonusers (Mariosa et al., 2020).
In summary, our cohort study found protective effects associated with hospital-based diagnoses of type 2 diabetes and obesity on ALS risk. The effects were most clearly observed among men and those aged 60 years or older. Despite these protective effects, the absolute rate differences were small, preventing clear clinical implications of these findings. Findings add to existing knowledge that energy metabolism plays a role in ALS occurrence.

AUTHOR CONTRIBUTIONS
All authors contributed to the design of the study. EHP and HTS acquired the data. All authors directed the analyses, which was carried out by SKS. NS and EBR wrote the initial draft. All authors contributed to the discussion and interpretation of the results, which secured the intellectual content of the manuscript. All authors accepted the final version for submission.

CONFLICT OF INTEREST STATEMENT
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 has any relation to the present study.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from Statistics Denmark.