The Malmö Diet and Cancer Study is a large population-based prospective study from Malmö, a city in southern Sweden . During 1991–1994, a random sample (n=6103) of participants from the Malmö Diet and Cancer Study were invited to participate in a sub-cohort study, the Malmö Diet and Cancer Study Cardiovascular Cohort . Of these, 4658 participants had complete data on prostasin and covariates, and the cross-sectional association between prostasin and diabetes was assessed in these participants (age 57.5 ± 5.9 years, 39.9% men) (Electronic supplementary material (ESM) Fig. 1). We then excluded 361 participants with prevalent diabetes, defined as self-reported diabetes or use of glucose-lowering medication or fasting venous blood glucose concentration ≥6.1 mmol/l (corresponding to a fasting plasma glucose concentration cut-off of 7.0 mmol/l ). Therefore, 4297 participants remained for the prospective analysis (age 57.3 ± 5.9 years, 38.5% men), including 232 with a history of cancer and 93 with cardiovascular diseases at baseline. Information on insulin levels was available for 4627 and 4247 participants in the cross-sectional and prospective analyses, respectively.
All participants provided written informed consent. The study conformed to the ethical guidelines of the 1975 Declaration of Helsinki, and was approved by the Lund University Ethical Committee (LU51/90, LU 2009/633, LU 2011/537 and LU 2012/762).
Information regarding smoking habits, alcohol consumption, current medication, leisure-time physical activity and educational level was obtained from the questionnaire. Participants were classified as smokers or non-smokers (including former smokers and never smoker). A daily alcohol intake >40 g for men and >30 g for women was considered to be a high level of alcohol consumption. Physical activity was represented by an overall leisure-time physical activity score calculated by multiplying the duration of specific activities by the corresponding intensity coefficient. Education was categorised as elementary or less, primary, secondary, upper secondary and further education without a degree and with a university degree. Waist circumference (cm) was measured midway between the iliac crest and the lowest rib margin. BP (mmHg) was measured using a mercury-column sphygmomanometer after a 10 min rest in a supine position. Fasting blood glucose levels (mmol/l) were measured from fresh plasma samples collected from the cubital vein after an overnight fast, according to standardised procedures, at the Department of Clinical Chemistry, Skåne University Hospital. LDL-cholesterol levels (mmol/l) were estimated using Friedewald’s formula. Fasting plasma insulin (mU/l) was measured using radioimmunoassay, and values were converted into pmol/l using the conversion factor 6.945. HOMA2-IR was calculated using a HOMA2-IR calculator . C-reactive protein (CRP, mg/l) was analysed using a Tina-quant CRP latex assay (Roche Diagnostics, Switzerland). Renal function, as reflected by eGFR, was estimated from plasma creatinine and cystatin C as previously described . Plasma creatinine was assessed by the Jaffé method using a Beckman Synchron LX20-4 (Beckman-Coulter, USA), and plasma cystatin C was assessed using a particle-enhanced immunonephelometric assay (N Latex Cystatin; Dade Behring, USA). The plasma prostasin concentration was determined using the Proseek Multiplex Oncology I v2 96 × 96 panel (UniProt ID Q16651; Olink Proteomics, Sweden) in blood samples that had been stored at −80°C after collection at baseline. The analyses were performed by SciLifeLab Laboratories (Uppsala, Sweden; www.ScilifeLab.se), and the same batches of reagents were used for sample assays. Olink NPX Manager Software was used for quality control of the samples and technical performance of the assays. During this process, four internal controls were added for all samples, and external controls were used in every analysis. The lower and upper limits of quantification of prostasin were approximately 0.24 and 7800 pg/ml, respectively. The potential impact of haemolysis on prostasin quantification was evaluated by using serial dilutions of haemolysate (0.25–15 mg/ml) in EDTA-treated plasma. The highest concentration of haemolysate (15 mg/ml) had no impact on assay performance. The intra-assay (within a run) and inter-assay (between runs) coefficients of variation were 5% and 19%, respectively. A pre-processing normalisation procedure was used to normalise technical variation within one run and between runs, by subtracting the Cq value for the extension control and the interplate control for each assay, respectively. The data were set relative to a correction factor, and are expressed as normalised protein expression (NPX) values on a log2 scale. The Olink webpage (http://www.olink.se) provides detailed information regarding proteomic panels, proteomics extension assay technology, assay performance, quality control and validation.
For analyses using diabetes as the endpoint, participants free of diabetes at baseline were followed until incident diabetes, emigration from Sweden, death or the end of follow-up (31 December 2019), whichever came first. Incident cases of diabetes were identified by linkages to both local and national registers, and have been described in detail previously . In brief, information was retrieved from six sources (the Swedish National Diabetes Register, the Regional Diabetes 2000 Register of the Scania Region, the nationwide Swedish drug prescription register, the Swedish inpatient register, the Swedish outpatient register and the Malmö HbA1c Register). In the Swedish National Diabetes Register and the Regional Diabetes 2000 Register, diabetes was diagnosed according to established criteria (fasting plasma glucose concentration ≥7.0 mmol/l with two repeated tests on separate occasions). In the nationwide Swedish drug prescription register, a filled prescription of insulin or glucose-lowering medications (ATC code A10) was required for diagnosis of diabetes. In the Swedish inpatient and outpatient registers, diabetes was diagnosed by a senior physician according to the established criteria. In the Malmö HbA1c Register, individuals were considered to have developed diabetes if they had at least two HbA1c recordings ≥42 mmol/mol (≥6.0%) using the Swedish Mono-S standardisation system (corresponding to 53 mmol/mol [7.0%] according to the US National Glycohemoglobin Standardization Program).
The Swedish cause of death register was used to monitor cause-specific mortality. Cancer mortality was defined as ICD-9 codes 140–239 (http://www.icd9data.com/2007/Volume1/default.htm) or ICD-10 codes C or D00–D48 (http://apps.who.int/classifications/icd10/browse/2016/en), and cardiovascular mortality was defined as ICD-9 codes 390–459 or ICD-10 codes in chapter I, as underlying cause of death.
Baseline characteristics of the study population are presented across quartiles of prostasin, with sex-specific quartile limits. Continuous variables with a skewed distribution are presented as medians (IQR) and were natural-logarithmically transformed before analyses, otherwise results are presented as means ± SD. Categorical variables are presented as numbers and percentages. Differences across quartiles were tested by linear regression for continuous variables and by logistic regression for categorical variables.
The cross-sectional association between prostasin and diabetes was analysed by multivariable logistic regression using three models. Model 1 assessed the crude association. Model 2 was adjusted for age, sex and waist circumference. Model 3 was additionally adjusted for smoking and drinking habits, LDL-cholesterol, systolic BP and anti-hypertensive medication. After excluding prevalent diabetes, the associations between prostasin (as the dependent variable) and baseline fasting blood glucose levels, plasma insulin levels and HOMA2-IR were assessed by multivariable linear regression after adjusting for the above-mentioned confounders. The association between baseline prostasin and incident diabetes was estimated using a Cox proportional hazards regression model, with time-on-study as the timescale. HRs and 95% CIs were calculated. Prostasin was analysed both as a continuous variable (per 1 SD) and grouped variable (quartiles). Potential covariates taken into account were age, sex, waist circumference, smoking and drinking habits, LDL-cholesterol, systolic BP and anti-hypertensive medication. Fasting blood glucose levels or HOMA2-IR were further adjusted for in a separate model, that may be considered as an overadjusted model. Other sensitivity analyses were performed, including additionally adjusting for CRP, eGFR, physical activity and educational level or lipid-lowering drugs, or adjusting for BMI instead of waist circumference in the final model. Possible interactions between prostasin and covariates in relation to diabetes risk were tested by incorporating interaction terms in the multivariable model. The proportional hazard assumption was tested by evaluating the time-dependent effects of prostasin on diabetes risk, and was also visualised using Kaplan–Meier curves. The linearity assumption of the association was examined using restricted cubic spline functions, with knots placed at 20%, 40%, 60% and 80% of prostasin concentration. We also analysed the association between prostasin (both by sex-specific quartiles and by SD) and mortality (all-cause mortality, cardiovascular mortality and cancer mortality) after adjusting for confounders including prior cardiovascular diseases or cancer, where appropriate. Sensitivity analyses were then performed after excluding prior cardiovascular diseases or cancer. The interaction between prostasin and fasting blood glucose levels was evaluated in relation to mortality risk. If an interaction was detected post hoc, analyses were separately performed after dividing participants according to their blood glucose level. Possible effect modifications by competing risk of death were explored using the Fine–Gray proportional sub-distribution hazards models method . C-statistics were calculated to estimate the added predictive value of prostasin to the multivariate model. Statistical significance was accepted with p values <0.05 (two-sided). All analyses were performed using SAS version 9.3 for Windows (SAS Institute, USA).