Study Population

The Vorarlberg Health Monitoring & Prevention Program (VHM&PP) is a population-based risk factor surveillance program in Vorarlberg, the westernmost province of Austria. The program is administrated by the Agency of Social and Preventive Medicine (aks). All adults (aged ≥19 years) within the province were invited to participate. Enrolment is voluntary and costs for one examination per year are covered by the participant’s compulsory health insurance. The screening examinations take place in the practices of local physicians according to a standard protocol. Anthropometric measures were carried out with participants wearing light indoor clothes and no shoes. Most of the participants have two or more registered visits with varying time interval between them. Details of the programme and characteristics of the study population have been previously described [4], [9], [24]. Between January 1985 and June 2005, 185,316 adult Vorarlberg residents were enrolled in the VHM&PP Study Cohort after signing informed consent. This corresponds to more than 60% of the general population in the eligible age range.

In order to assess two subsequent BMI change intervals, three measurement time points (t) were defined: baseline (t₀), year 5 (t₅), and year 10 (t₁₀) (Figure 1). For t₅ or t_10, data within the year after or the year before wereaccepted. Change of BMI was calculated between t₀ and t₅ (t_0–5) and the subsequent time interval t₅ and t₁₀ (t_5–10). In total, 43,233 men and women with complete data on height, weight and smoking status were available. No information was available regarding participant morbidity status at baseline. Therefore, due to increased probability of pre-existing disease, participants who were underweight (BMI<18.5 kg/m²) at baseline were excluded (n = 1,134).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Time line of BMI assessments, BMI change intervals and subsequent follow up.

https://doi.org/10.1371/journal.pone.0084817.g001

Classification of BMI at Baseline and Changes Over Time

Height and weight were measured by medical staff during the VHM&PP physical examinations. BMI at t₀, t₅, and t₁₀ was classified according to World Health Organisation (WHO) guidelines normal (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), and obese (≥30.0 kg/m²). Normal BMI was defined as reference category for statistical analyses. BMI change per year was calculated for each of the considered time intervals. Three categories were defined: weight loss (<−0.10 kg/m²/year), stable weight (−0.10 to 0.09 kg/m²/year), and weight gain (≥0.10 kg/m²/year). To analyse the patterns of two subsequent BMI change intervals, 27 different combinations of baseline BMI and BMI change were considered. People with normal weight and no weight change during both intervals was used as reference category.

Endpoint

All-cause mortality served as endpoint for all analyses. Follow-up started after the last BMI measurement at t₁₀. Death information was provided by the Vorarlberg Death Registry and was linked to the cohort data. All deaths were identified from death certificates and were confirmed by trained physicians only. In case of unclear causes of death, autopsies were performed (23.3% of all deaths in Austria in 2002) [25]. Observation time started 01.01.1985 and participants were censored by December 31, 2009.

Covariates

All analyses were adjusted for smoking by including information on current smoking and former smoking in the model. Persons with missing smoking data were classified as never smokers because the baseline questionnaires did not differentiate between never-smokers and participants with missing values [9]. Smoking information from follow-up visits was available for most of the participants to complement the baseline smoking status. The category of never-smokers contains therefore only a low percentage of misclassified smokers and ex-smokers. Due to the potential for over-adjustment, we did not adjust for assumed intermediate steps in the causal pathway [11].

Statistical Analysis

Hazard rate ratios (HR) and 95% confidence intervals (95% CI) were used to estimate the effect of baseline BMI, combinations of baseline BMI and the first BMI change interval (t_0–5), as well as, combinations of baseline BMI and both subsequent BMI change intervals (t_0–5+ t_5–10) on all-cause mortality. Stratification according to baseline BMI was performed in order to explore possible effect modification.

Furthermore, the pattern of BMI over time was assessed for each participant by comparing changes in BMI category (as defined by the WHO) from t₀ to t₅ to t₁₀. (e.g. normal weight – overweight – normal weight). For t₅ and t₁₀ the category underweight was included. The relationship of these patterns with all-cause mortality was estimated by Cox proportional hazards models and was displayed as Hazard ratios (HR).

All analyses were performed separately for men and women. Models included age (in single years) in the strata statement and adjustment for smoking status. The calculations were performed using SAS 9.2.

Principal Findings

With the large number of participants with repeated anthropometric measurements and the reliable link to the local death registry we were able to study the association between BMI change during two subsequent time intervals of five years and all-cause mortality. Overall, the lowest risks were seen in normal weight and overweight subjects with stable BMI over time. Weight gain was associated with mortality in overweight and obese persons, although not statistically significant in all cases. Weight loss increased HRs in men and women consistently.

Our findings concerning the association between BMI at baseline and all-cause mortality are consistent with a previous analysis of a larger sample from the VHM&PP cohort [9] and with those from other large cohorts [6]–[8], [10], [11] This indicates that selection bias in the present subsample is unlikely and that the baseline mortality risk of BMI is comparable with other populations.

Overall, long-term weight stability was prevalent in 30% of our cohort. Our finding that stable BMI in normal weight and overweight participants showed the lowest risk for all-cause mortality is consistent with results of a recent review [26]. A stable BMI may reflect a more balanced lifestyle with less exposure to unhealthy behaviors such as smoking or physical inactivity [27]. However, weight gain was the most prevalent pattern in our cohort. A stable BMI during both time intervals was seen only in about 10% of all participants with normal weight at baseline. The proportion became considerable smaller by increasing BMI category. Similar findings were seen in the literature [26].

The observed positive relationships between weight loss and increased mortality in our study is consistent with several previous studies [12]–[18] However, the interpretation is not straightforward. Most studies accounted for health status, but information may have been incomplete or underlying diseases might not have been detected [18]. Furthermore, the majority of investigations did not differentiate between intentional and unintentional weight loss [18]. In our cohort, there was also no information about morbidity or intention of weight loss. However, in order to investigate the potential for unintended weight loss due to pre-existing diseases, we conducted sensitivity analyses excluding death three years after the last BMI measurement. The patterns did not change substantially (data not shown). In addition, we analysed weight loss with more detail than previous studies by investigating two consecutive time intervals of 5 years. Consequently, we observed that weight loss was associated with higher mortality regardless of weight change in the other time period. Nevertheless, our findings related to weight reductions and all-cause mortality should be interpreted with caution. Overall, long-term unintentional weight loss appears as an important health indicator, which needs to be further explored in age and disease specific context.

We found that continuous weight gain over both time intervals (t_0–5+ t_5–10) was associated with increased all-cause mortality in overweight and obese persons, although not statistically significant in men. The majority of previous studies also reported a positive relationship between increase in body weight and mortality [12], [13], [15], [19]–[21], however, several studies did not find any relation [14], [16], [17] The varying results may be due to different approaches for assessing weight change. Weight and height were either recalled from the past, self-reported or measured. Our study used only valid height and weight measurements assessed by trained medical staff. Furthermore, the number of measurements and the length of follow-up varied considerably between studies. Data from the Framingham heart study showed that the length of follow-up influenced the association of BMI and mortality [28].

Another issue concerning the literature on weight change and mortality is the lack of accepted cut-points to define categories of weight change. Consequently, weight change categories differed between studies. In some studies, the highest category of weight gain included participants with only moderate weight gain (≥5 kg within 10 years) [12], [14] Conversely, in one North-American investigation, the highest category of weight gain comprised women gaining more than 40 kg within a varying assessment period of 12–37 years [15].

In contrast to many other studies that define absolute weight gain as the primary outcome, our study benefits from the use of relative measures (change in kg/m²/year). Weight stability in our categorization refers to ±2.6 kg in a person of 1.60 meters of height and ±3.4 kg in a person of 1.85 meters over a period of 10 years and for weight gain to ≥12.8 kg and ≥17.1 kg, respectively.

Strengths and Weaknesses of the Study

A major strength of our study is that we could analyze long-term BMI change patterns of two subsequent 5-year-intervals on an individual level in a very large study population with more than 10 years of follow-up. Additionally, we used widely accepted BMI categories specified by the WHO whereby results can be used for public health recommendations. Finally, as the average age within our study population was 43 years, our cohort should have consisted of few chronically ill individuals. Therefore, we believe that unintentional weight loss as result of chronic illness was minimal.

Using BMI as a measure of overweight and obesity may be considered as a limitation. BMI reflects different body compartments and is not informative concerning the fat and lean body mass distribution [29]. Smoking affects the BMI and the body fat distribution [30], which may have confounded the relationship between BMI change and over-all mortality. However, exclusion of ever smokers revealed similar patterns, but increased estimates in obese never-smoking men (Table S1 and S2 in File S1). The ratio between fat and lean body mass is a result of physiology, aging, energy intake, and physical activity [31], [32] Aging is characterized by loss of subcutaneous fat and ectopic fat deposition [31]. However, due to the high correlation of BMI with body fat, BMI is generally considered a reasonable measure of obesity in epidemiological studies. We used BMI between 18.5 and 24.9 kg/m² as reference category. A recent meta-analysis including 2.88 million individuals showed no mortality increase for BMI values less than 35 kg/m² [11]. This is consistent with our finding that weight stability in overweight persons did not increase mortality risk. There could be an age-dependent range of BMI levels associated with equivalent risk of death [33]. Thus, more refined measurements of the fat-lean body mass ratio may be necessary to further clarify the association between BMI and mortality in relation to age [7]. Finally, despite the large sample size results did not reach statistical significance in some cases. However, dose-response relationships were seen. Unfortunately we do not have information on further potential risk factors such as comorbidities. Concerning generalizability it has to be noted that participants voluntary participated in the study and had to survive at least 10 years due to our study design. Therefore, mortality of this cohort might be lower compared to the general population.

Conclusions

In summary, our data indicate that obesity should be avoided and that weight gain in overweight and obese persons is a risk factor for all-cause mortality. The results regarding weight loss and mortality, though consistent in all time intervals, should be interpreted with caution. The lowest mortality risks were seen with normal weight and overweight people maintaining BMI over 10 years. Therefore, our findings highlight weight stability within these BMI ranges as an important public health prevention strategy.

Ethical Approval

Ethical approval the Vorarlberg ethic commission EK-Nr. 2006-6/2.