Sagittal Abdominal Diameter as a New Predictor for Incident Diabetes

This study is based on data from Health 2000, a health examination survey with a nationally representative sample of 8,028 people aged 30 years or older. The sample was drawn from 80 health service districts throughout Finland using a two-stage cluster sampling (28,29). Of the 6,771 people who participated in the health examination, those 5,590 who were aged 30 to 79 years and did not have diabetes at baseline were included in the current study. After further exclusion of pregnant women and individuals with missing information for the variables used in the analysis, the final dataset comprised 5,168 individuals (2,399 men and 2,769 women). The field phase was conducted in 2000–2001, and the individuals were followed up in a cohort study design for 8.1 years (41,689 person-years of follow-up). During the follow-up, type 2 diabetes developed in 222 individuals.

The field phase included a health examination, a home interview, and questionnaires. The study was approved by the ethics committee of the Hospital District of Helsinki and Uusimaa, and all participants gave a written informed consent (28,29). The methods used have been described earlier in detail and are briefly summarized here (29).

Weight was measured in lightweight clothes to the nearest 0.1 kg and height to the nearest 0.5 cm. BMI was calculated as body weight (kg)/height squared (m²). Waist girth was measured midway between the lower rib margin and the iliac crest, and the top layer of the lower body was undressed. SAD was measured using an instrument developed by the technical unit of the National Health Institute. The subject was asked to lie down, with the upper body naked, on the examination table. One of the wings of the measuring device was placed under the subject’s back, at the high point of the iliac crest. The subject was asked to relax and breathe regularly. The other wing of the measuring device was lowered gently onto the subject’s stomach. The measurement was performed after normal expiration. The intraclass correlation coefficient for the raters’ repeatability of the SAD measurements was 0.88, and the agreement between raters was of the same order.

Data on education, smoking, leisure time physical exercise, alcohol consumption, previous diseases, and medication were self-reported in a health interview or a self-administered questionnaire at baseline. Casual blood pressure was measured after a 5-min rest twice at a 1.5-min interval using a standard mercury sphygmomanometer, and the mean of the two measurements was used in the analyses. Participants were asked to fast for at least 4 h before the examination. Blood samples were drawn and stored at −70°C. Serum HDL cholesterol, triglycerides, and glucose levels were determined as soon as technically possible (usually some weeks) after the samples were taken. Glucose was measured using a hexokinase method (Olympus System Reagent, Hamburg, Germany), and HDL cholesterol (HDL-C Plus, Roche Diagnostics GmbH, Mannheim, Germany) and triglycerides (GPO PAP, Olympus System Reagent) were measured using a photometric enzymatic method.

The harmonization definition (12) was used for defining the metabolic syndrome. A positive definition requires three or more of the following five criteria to be satisfied: large waist girth (≥94 cm in men and ≥80 cm in women), hypertriglyceridemia (≥1.7 mmol/L), low HDL-cholesterol level (<1.0 mmol/L in men and <1.3 mmol/L in women), elevated blood pressure (systolic ≥130 and/or diastolic ≥85 mmHg) and/or antihypertensive drug treatment or history of hypertension, and elevated fasting serum glucose (≥5.6 mmol/L) or antidiabetic drug treatment.

Nationwide administrative registers, which reliably cover every death, hospitalization, and medication use, were used to find cases of incident diabetes by linking the participants of the current study to these registers by a unique personal identification code assigned to each Finnish citizen (5). The identification was based on the presence of any of the ICD-10 codes E10 to E14 in at least one of three registers: the Drug Reimbursement Register (30) and covering all patients with diabetes receiving free hypoglycemic drugs under the Sickness Insurance Act; the Finnish Hospital Discharge Register (31); and the National Causes-of-Deaths Register (32). These data were augmented with information on the purchase of antidiabetic drugs (Anatomical Therapeutic Chemical codes beginning with A10) by individuals who had been diagnosed with diabetes but had not yet received a reimbursement decision according to a nationwide Prescription Register maintained by the Social Insurance Institution.

The Cox proportional hazards model was used to calculate the relative risk (RR) for developing incident diabetes during the follow-up. The follow-up time was defined as days from the baseline examination to the date of type 2 diabetes occurrence, death, or end of follow-up, whichever came first. Because of the two-stage cluster sampling design, the analyses were conducted using the program package SUDAAN, release 10, using the Taylor series linearization method and poststratification weights (33). A piecewise constant hazards model was used to assess the PAF for the three obesity indicators (34). The PAF estimates the proportion of cases in a given population that would theoretically not have occurred if all the individuals had had low-risk target values of the risk factors of interest instead of their true values. This is performed by combining information about the prevalence of the risk factor in the population with estimates of the strength of the association between the risk factor and the outcome. The PAF analyses were performed using an SAS macro (35,36).

Four main effect models were used. The first model included age, sex, education, smoking, alcohol consumption, physical exercise, and one obesity indicator at a time. The second model was otherwise identical to the first model but included all three obesity indicators. In the third and fourth models, serum triglycerides, HDL-cholesterol, glucose, and blood pressure were entered into the first and second models.

Possible interactions between the obesity indicators were also studied in four models. The first model included age, sex, education, smoking, alcohol consumption, physical exercise, and interaction terms of two obesity indicators at a time. The second model was otherwise identical to the first model but also included the third obesity indicator. The two remaining models were formed by entering serum triglycerides, HDL cholesterol, glucose, and blood pressure in the former two models. These eight models were used for estimation of the RR and the PAF.

The mean value of the SAD was 21.5 cm (SD 3.19). Individuals in the highest quartile were more likely to be men, to have a lower educational attainment, to smoke less frequently, to consume more alcohol, and to exercise less than those belonging to the lower quartiles of SAD (Table 1). The likelihood of harmonization metabolic syndrome increased with increasing SAD (P < 0.001 for trend, Table 1)

Table 2 displays the adjusted RRs of diabetes in quartiles of SAD, BMI, WHR, and waist girth, and the corresponding PAF. All indicators predicted a more than 10-fold risk of diabetes after adjustment for age, sex, education, smoking, alcohol consumption, and physical exercise (model 1). The RR (95% CI) of diabetes between the highest and lowest quartile was 14.7 (6.89–31.2) for SAD, 15.0 (6.94–32.6) for BMI, 11.4 (5.39–23.8) for waist girth, and 12.5 (6.47–24.2) for WHR. The corresponding PAF estimates (95% CI) were 83% (67–92), 85% (69–93), 79% (61–88), and 78% (63–87), indicating that approximately four of five diabetes cases could have been avoided if the obesity indicators had been in their lowest quartile. After simultaneous inclusion of all three obesity indicators in the model, the RRs were considerably attenuated (model 2). Adjustment of models 1 and 2 with components of the harmonization metabolic syndrome further attenuated the RRs (models 3 and 4). The same was true for the PAF estimates. A total of 61% (19–81) of all diabetes cases could have been avoided if the SAD of all subjects had been in the lowest quartile (model 4). The corresponding values were 60% (15–82) for BMI, 46% (−2 to 72) for waist girth, and 46% (7–68) and for WHR.

RRs of diabetes generated by pairwise interaction terms of the three obesity indicators are reported in Table 3. Of the combinations, the co-occurrence of high BMI and high SAD was associated with the highest diabetes risk. The RR (95% CI) between the highest and lowest quartiles was 37.0 (11.2–122) in model 1, and further adjustment for waist girth and the components of the harmonization metabolic syndrome (model 4) attenuated the RR to 9.88 (2.81–34.7). Adjustment affected the PAF estimates (95% CI) of the BMI and SAD combination only marginally, being 93% (78–98) in model 1 and 84% (49–95) in model 4. The predictive power of the interaction between waist girth and WHR and SAD was moderate, and the interaction between BMI and waist girth proved the least predictive of the combinations (PAF 50% [11–71]). Thus, 50% of the diabetes cases could have been avoided if BMI was <25 kg/m² and waist girth was <94 cm in men and <80 cm in women, whereas 84% could have been avoided if BMI was <25 kg/m² and SAD was <20 cm after adjustment for the other factors of the metabolic syndrome.

We found that SAD, BMI, WHR, and waist girth all are powerful predictors of incident diabetes during an 8.1-year follow-up period. Our data further revealed that combining information on SAD and BMI outperforms information gained by measuring SAD, BMI, or waist girth alone, or any other combination thereof. Indeed, even after adjustment for age, sex, and lifestyle factors, the combination of high SAD and high BMI showed a nearly 37-fold increased risk of diabetes incidence compared with the risk of individuals who had normal BMI (<25 kg/m²) and belonged to the lowest SAD quartile (13.5–19 cm). Even when the components of the harmonization metabolic syndrome were taken into account, the risk associated with co-occurring high SAD and high BMI remained nearly 10 times higher. The corresponding PAF was 84% compared with 61% and 60% for the individual components of the combination, respectively. The population-level importance of observing and attempting to influence both of these aspects of obesity in clinical work would thus be expected to be significant.

To our knowledge, this is the first longitudinal study comparing SAD and other anthropometric measures in predicting diabetes incidence. The results support earlier findings (4) indicating that not only the degree of obesity but also the location of fat is a risk factor for diabetes. SAD correlated strongly with visceral obesity in normal-weight and obese individuals (15–18). When SAD is measured in a supine patient, the loose subcutaneous fat falls toward the sides due to gravity, leaving harder visceral fat to contribute to SAD measurement. Our findings are in line with results showing that SAD is associated with diabetes in a cross-sectional setting (26) and is a predictor of incident cardiovascular disease (24,25), even when SAD is measured with the subject standing (24).

BMI is an overall measure that does not differentiate lean mass from adipose tissue, and waist girth does not differentiate between subcutaneous fat and visceral fat. Thus, it is not surprising that a combination of SAD and BMI, both of which take into account overall fatness and indirectly estimate the amount of visceral fat, seems to be the most powerful risk indicator for diabetes.

In the current study, waist girth was the least powerful of the obesity indicators studied. In the multivariate Cox proportional hazards regression models, waist girth lost the most of its predictiveness when the models were adjusted for variables correlating with waist girth. Furthermore, in the interaction analyses, the models that included waist girth showed the weakest prediction. Of the obesity indicators, waist girth is the indicator that is currently incorporated in the newest definition of the metabolic syndrome. Although this harmonization definition does predict diabetes incidence in line with earlier definitions (37), the current study suggests that further studies are needed to assess the most powerful obesity measure to be used in the future metabolic syndrome definitions and diabetes risk algorithms.

Before SAD can be more widely used in the clinical setting, properly validated measurement protocols and evaluation of cutoff values are needed. Study of receiver operating characteristic curves did not, however, reveal any clear cutoff point, suggesting that the diabetes risk changes over the whole range of SAD.

In the current study, SAD identified normal-weight high-risk individuals: those with a BMI of <25 kg/m² but with a large SAD were at increased risk of diabetes (data not shown). However, the number of individuals with a low BMI and a high SAD was low, and thus, the public health value of this particular finding would appear to be marginal. A larger study population is needed to clarify the importance and meaning of this finding.

The strengths of the current study were its prospective design, valid follow-up data, and the comprehensive set of methods used to measure body composition. The anthropometric measurements were performed by trained personnel who used a systematic protocol. SAD was measured using a noncommercial instrument developed by the technical unit of the National Health Institute, and the reliability of the SAD measurements was high. Furthermore, we used the most recent consensus definition of the metabolic syndrome (12).

Despite the advantages, some limitations should be kept in mind. Our study sample was nationally representative for Finland. However, the Finnish population mainly consists of northern European Caucasian individuals, and thus, our results should be replicated in other types of populations. The number of events was relatively low in some of the subcategories of the cohort, which might have influenced the results. Because of the small number of subjects in the reference group, it was only possibly to do the analyses with men and women combined. Our outcome was clinically incident diabetes, and we did not obtain glucose tolerance tests at the end of the follow-up. Thus, we were only able to identify individuals who received drug treatment for diabetes or were hospitalized or died during the follow-up, but we could not identify undiagnosed individuals or diabetic persons treated with diet only. This means that the incidence of diabetes was underestimated in the follow-up, which probably makes our estimates rather conservative.

In conclusion, we demonstrated in this longitudinal population-based study that the combination of SAD and BMI measurements may provide a new powerful predictor of incident diabetes in the general population. Because this is the first etiological study to demonstrate the ability of SAD to predict diabetes incidence, future studies have to confirm our finding until a firm conclusion about screening can be made.

P.P. was supported by the Finnish Medical Foundation and is currently employed by Novo Nordisk Farma. No other potential conflicts of interest relevant to this article were reported.

P.P. wrote the manuscript. H.R. and M.A.L. researched data and reviewed and edited the manuscript. M.H., A.R., and P.K. reviewed the manuscript and contributed to discussion. H.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.