Long-term Mortality in a Cohort of Severely Obese Persons in Italy
Few large studies on Northern European or US populations reported on mortality of severely obese individuals (BMI ≥ 40 kg/m2). We studied a historical cohort in Italy to compare its mortality with previous findings, to investigate its relationship with BMI in the >40 range, and to provide evidence useful for clinical decision-making on treatment. The cohort comprised 4,837 persons with a BMI ≥40 kg/m2 and aged ≥18 at first consultation, referred to six centers for obesity treatment between 1975 and 1996. After exclusion of persons with missing personal identification data or those untraceable, 4,498 (972 men, 3,526 women) remained for analyses. We calculated standardized mortality ratios (SMRs) and carried out Cox proportional hazards modeling. General mortality (484 deaths: 153 men, 331 women) was in excess, with SMRs (95% confidence intervals) of 2.78 (2.36–3.26) for men and 2.10 (1.88–2.34) for women. Excess mortality (i) was observed in all BMI categories, except among women weighing 40–42.4 kg/m2; (ii) increased with increasing BMI; (iii) increased less among persons recruited in recent calendar periods; (iv) was inversely related to age attained at follow-up; and (v) was due to cardiovascular and respiratory diseases and violent deaths but not malignant neoplasms. Excess mortality was similar to that observed in Northern European and US cohorts. Its steady increase with BMI levels ≥40 suggests that benefits proportional to weight reduction are expected and that even limited control may be beneficial. The smaller excess among persons recruited most recently might reflect better treatment.
The association of obesity with increased mortality and morbidity is firmly established. Prospective cohort studies conducted in the United States and Northern Europe have shown that, in the long term, mortality increases with BMI (body weight in kilograms divided by the square of height in meters), starting at BMI values >20–24 kg/m2 (1,2,3,4,5,6,7,8,9,10).
In severe, or morbid, obesity (i.e., BMI ≥ 40 kg/m2) mortality was two to three times higher than expected (1,6,9). Epidemiological studies on this high-risk group in Italy are limited (11,12), showing a standardized mortality ratio (SMR) lower than that expected from observations in non-Mediterranean populations (12). There is interest, therefore, in assessing whether the mortality pattern already associated with morbid obesity can be confirmed in different geographical, socioeconomic, and nutritional contexts.
Trends in mortality with increasing BMI values have never been studied above the threshold for morbid obesity, and understanding how mortality among severely obese individuals is affected by BMI would be useful for guiding clinicians in decisions regarding treatment. As treatment itself can entail risks, the more precise the risk-benefit profile at the individual level the better (13).
In order to investigate the general and cause-specific mortality of severely obese persons, to assess possible differences in health risks between Italian and Northern European populations, and to study the relationship between mortality and BMI level in the range of severe obesity, we set up and followed a large historical cohort of patients seeking treatment at six Italian centers specializing in obesity.
Methods and Procedures
Our records covered 4,837 individuals. Inclusion criteria were first attendance at any of the recruiting centers between 1975 and 1996, age ≥18 years at first visit and BMI at first visit ≥40 kg/m2. Although six centers collaborated in the enrolment of this cohort, most persons were seen at the Istituto Auxologico Italiano Hospital in Piancavallo.
Some persons were excluded from follow-up, because of duplicate registration, incomplete data for personal identification, missing age or BMI at first visit, missing information on the municipality of last known residence, leaving 4,681 subjects in the cohort. For every cohort member, the observation period (follow-up) started at the date of first visit at a participating center (date of enrolment) and the vital status was ascertained by contacting the town hall of the municipality of his or her most recent residence. People who had moved were traced by repeating the inquiry at their new municipality of residence until emigration, death (and place of death), or vital status after the conventional follow-up closing date, 31 December 2001, was ascertained. Death certificates, as filled out by the attending physicians, were obtained from the local health authorities of the municipalities where the deaths had occurred. Cohort members contributed to the person-years calculation until emigration, death, or end of follow-up, whichever came first. Causes of death were coded according to the International Classification of Diseases, 9th revision (ICD-9).
SMRs were computed separately for the two genders, for all causes (general mortality), and for the following specific causes and groups of causes: all malignant neoplasms (ICD-9 codes 140–208), tumors of the esophagus (150), colon (153), rectum (154), pancreas (157), lung (162), breast (174,175), and kidney (189), non-Hodgkin's lymphomas (200,202), multiple myeloma (203), leukemias (204–208), all cardiovascular diseases (390–459), hypertension (401–405), ischemic heart diseases (410–414), acute myocardial infarction (410), all respiratory diseases (460–519), chronic obstructive pulmonary diseases (490–496, 416), bronchitis (490), chronic bronchitis (491), emphysema (492), asthma (493), diabetes (250), obesity (278.0), and violent deaths (800–999). General and cause-specific mortality rates for the general Italian population for the two genders, in 5-year age classes and 5-year calendar periods, were used to compute the expected number of cases. Stratified analyses were carried out according to the BMI recorded at the first visit and according to the calendar period of enrolment. Person-years of observation, SMRs, and 95% confidence intervals were computed using the statistical package OCMAP 3.10 (University of Pittsburgh, Pittsburgh, PA) (14).
The homogeneity of the SMRs and their trends in the stratified analyses were tested according to the method suggested by Breslow and Day and by the Poisson test statistic, respectively (15). Multivariate analyses for total mortality were conducted by fitting Cox regression models in Stata 9 for Windows (Statacorp, College Station, TX). The Cox regression models were adapted by using age as the timescale for analysis, the calendar period of observation (5-year periods) as the stratification variable, and gender, BMI at recruitment (BMI classes: 40–42.4, 42.5–44.9, 45.0–49.9 and ≥50.0 kg/m2) and calendar period of recruitment (5-year periods) as predictive variables. The likelihood-ratio test was used to assess improvements in data fitting in models of increasing complexity.
The general characteristics of the cohort members are shown in Table 1. Vital status could not be ascertained for 183 cohort members of the 4,681 (3.9% lost to follow-up), leaving 4,498 persons for the analyses. The distribution of person-years by age group and calendar period is shown in Table 2. A large proportion of person-years were accrued in age groups up to 50–59, comprising 25,706 (73.4%) in women and 7,972 (85.4%) in men.
Table 1. General characteristics of cohort members
Table 2. Person-years by gender, age, and calendar period of follow-up
Of the 517 deaths observed, 484 (153 men and 331 women) occurred before the end of follow-up, giving SMRs of 2.78 (95% confidence interval, 2.36–3.26) for men and 2.10 (95% confidence interval, 1.88–2.34) for women. Table 3 shows mortality according to BMI at first visit. Among women a clear-cut monotonic increase in mortality with increasing BMI was found (test for trend of SMRs, χ 2 = 50.30, P < 0.001), whereas for men the pattern was less regular (test for trend of SMRs, χ 2 = 6.61, P < 0.025). Table 3 also shows mortality by calendar period of enrolment based on the date of first attendance at a recruiting center. In both genders, mortality decreased in the most recent periods (test for trend of SMRs, χ 2 = 7.90, P < 0.01 among men; χ 2 = 8.79, P < 0.01 among women). Individuals recruited in the most recent calendar periods had limited follow-up duration; should the consequences of obesity occur mainly after a long latency, they would not have been fully at risk of excess death and thus their SMRs could have been underestimated. To control for this possible bias, the analysis was repeated with the observation period limited to the first 5 years of follow-up, with no significant change in results (data not shown).
Table 3. Mortality from all causes by gender and by BMI at recruitment and calendar period of recruitment
Table 4 shows how mortality patterns changed with age. A limited number of person-years were accrued above the age of 60 in both genders, but especially among men, as can be seen in Table 2 and by the trends in expected deaths by age group in Table 4. Nevertheless, large numbers of excess deaths were observed, starting in early adulthood. More specifically, whereas most of the excess deaths occurred in late adulthood and among elderly persons, the highest age-specific observed/expected ratios were seen in early and full adulthood in both genders. There appeared to be a bimodal distribution, with the ratios peaking first at age 25–29 in both genders and then at age 45–49 among men and 55–59 among women.
Table 4. Mortality from all causes, by gender and age group in the study of severely obese persons in Italy
In multivariate analyses, a full Cox regression model (including gender, BMI, and period of enrolment) minimized the deviance and significantly improved data fitting according to the likelihood-ratio test results. The increase in mortality with increasing BMI at first visit was monotonic, as was the decreasing trend in the most recent recruitment periods (Table 5).
Table 5. Results of the Cox regression model including full parameters in the study of severely obese persons in Italy
Cause-specific mortality, limited to causes and groups of causes with at least five deaths in either gender, is shown in Table 6. Mortality from malignant tumors was not increased in either men or women. The only tumors with SMRs significantly higher than unity were kidney cancer in women and breast cancer in men (based on 1 observed death). Mortality from cardiovascular and respiratory diseases was significantly in excess in both genders, particularly that from hypertension, ischemic heart disease (in women), chronic obstructive pulmonary disease and, more specifically, bronchitis, emphysema, and asthma. The SMRs for obesity and diabetes (in men) were markedly increased. Deaths from violence were significantly in excess in men, but the excess for women did not reach statistical significance.
Table 6. Mortality from specific groups of causes of death, by gender, in the study of severely obese persons in Italy
We studied a group of severely obese persons who had sought treatment in specialized centers. In previously published studies, cohort members had been recruited to represent samples of the general population, in which subgroups with high BMI at enrolment were subsequently identified. They thus represented a limited fraction of the study population, and only in the largest cohorts were the numbers of severely obese persons of the same order of magnitude as ours. The American Cancer Society Cancer Prevention Study II was based on 1,315 person-years of observation in the group with a BMI ≥ 40 kg/m2 for men, with 19 deaths, and 14,372 person-years for women, with 144 deaths (1). In a Norwegian study, 8,119 person-years were accrued, with 269 deaths, in men with a BMI ≥ 40 kg/m2 at first visit and 116,713 person-years with 3,801 deaths in women (6). Therefore, this study is one of the largest ever conducted presenting findings on long-term mortality in morbid obesity.
In an Italian survey involving a nationwide random sample of 52,000 families and 140,000 individuals, the prevalence of obesity and severe obesity in adults was very similar in both genders, ∼9 and 1%, respectively (16). The high proportion of women in our cohort reflects attendance to the recruiting centers and differs from the expected equal representation of genders. However, it does not entail bias in risk estimates, as all analyses were conducted separately for men and women.
Our results show a large excess of mortality from all causes in both genders. The increase was almost threefold in men and more than twofold in women, and is similar to that observed in studies in North America and Northern Europe. In the cohort of the American Cancer Society Cancer Prevention Study II, in which height and weight were self-reported on a questionnaire completed at recruitment, the mortality of persons with a BMI ≥ 40 kg/m2 was considerably higher than that of persons in the reference category (BMI, 23.5–24.9 kg/m2) after 14 years of follow-up (1). The relative risk for mortality was 2.58 for white men and 2.00 for white women after adjustment for age, educational level, physical activity, and other confounders in multivariate analyses based on a Cox proportional hazards model. In the Norwegian follow-up study, in which height and weight were measured in a standardized way, and repeated individual measurements were available for some sub-cohorts, the mortality risk associated with a BMI ≥ 40 kg/m2 at first visit after 22 years of observation was 2.53 for men and 1.93 for women in comparison with the reference category (BMI, 22.5–23.5 kg/m2) (6). The relative risks were estimated by Cox proportional hazards modeling to allow for the effect of confounders such as age and birth cohort. In a German study of obese persons, general mortality in the group with a BMI ≥ 40.0 kg/m2 was increased, with relative risks of 3.05 for men and 2.31 for women (9).
A possible limitation of our cohort study is that it is based on persons seeking medical advice rather than on presumably healthy individuals. Therefore, we cannot consider it representative of all severely obese persons in Italy, and it might not be comparable with other studies. We could not overcome this problem by adjusting for comorbidity in our analyses, as we had no information on it. Nevertheless, our findings, with SMR = 2.78 in men and 2.10 in women, suggest that the strength of the association with mortality in this cohort is very close to the estimates from previous studies conducted in Northern Europe and in the United States (1,6,9) and is at variance with the only report from Italy (12), where a considerably lower estimate (1.61, both genders) for the SMR was obtained.
This is the first study in which mortality at increasing BMI values above the threshold for severe obesity was investigated. In multivariate analysis, mortality increased monotonically with BMI level and was extremely elevated in both men and women at BMI values ≥50 kg/m2. Adjustment for potential confounders such as smoking, physical activity, and socioeconomic status was not possible owing to lack of proper data. However, the association between BMI level and mortality is strong and could hardly be completely explained by uncontrolled confounding factors. The striking increase in mortality risk across the whole BMI range we investigated highlights the importance of strict weight control, but suggests that BMI reduction could be beneficial even if relatively limited and within the boundaries of morbid obesity. It is worth noting that, in univariate analysis, the SMR for women with BMI 40.0–42.4 kg/m2 was not significantly increased. In this group of patients, the choice of treatment requires, therefore, a careful balance of risks and benefits.
Mortality was statistically significantly increased in all calendar periods of enrolment for both men and women. The SMRs decreased considerably, however, in the most recent periods, by half among men in comparison with the first period, with a similar, although less accentuated, pattern in women. The same trend was found in the multivariate analysis. This observation is a unique feature of the design of our study and was perceptible owing to the continuous enrolment of participants >21 years. The reduction in excess mortality might reflect more effective treatment and better prevention of sequelae. All recruiting centers were medical units. We lacked information on subsequent treatments received by our cohort members and it is likely that some received bariatric surgery. However, while 200,000–400,000 severely obese individuals may exist in Italy today, ∼14,000 patients undergoing bariatric surgery 1996–2006 are present in the National Registry maintained by the Italian Society of Obesity Surgery (17). Therefore, surgical treatment for obesity is still relatively rare in Italy and we believe that the period effect we observed is unlikely to be due to surgical interventions. Thus, considerable room for further improvements seems to exist, as patients treated with bariatric surgery exhibited a striking reduction in mortality (18).
Cause-specific mortality could be analyzed only according to major groups and some selected causes of death, because of the limited number of observations. Increased mortality from cardiovascular and respiratory diseases and from violent deaths, but not from malignancies, was observed. Cardiovascular and respiratory diseases are frequent complications of obesity (19,20). In particular, the obesity hypoventilation syndrome is associated with morbid obesity, may cause complex and very severe impairment of pulmonary function, and may be misdiagnosed as chronic obstructive pulmonary disease (21). Mortality from chronic obstructive pulmonary disease and, more specifically, from bronchitis, emphysema, and asthma was greatly increased. The absence of excess deaths from cancer, both overall and site-specific (with the only exceptions of kidney cancer in women and breast cancer in men), may seem surprising, as obesity is firmly associated with some malignancies, including esophageal, colon, rectum, pancreas, breast, and kidney tumors as well as with non-Hodgkin's lymphoma, multiple myeloma, and leukemias (8,22). However, it must be taken into account that our study included relatively young individuals, so that only a limited proportion of person-years was accrued in age classes >60, where cancer mortality becomes substantial. Due to the limited number of observations, we were unable to estimate risks for most cancer sites of interest. Therefore, this cohort is at present uninformative on cancer risk and further follow-up is warranted. A striking increase in deaths due to obesity could be observed in both genders, on the other hand, accompanied by an elevated SMR for diabetes in men.
The harmful long-term potential of severe obesity is clearly shown by our findings, confirming that similar effects occur in different geographical, social, educational, and nutritional contexts in North America and in Northern and Southern Europe. They contribute thus to the public health alert concerning the worldwide increase in the prevalence of obesity. The steadily increasing trend in mortality according to BMI above the threshold of severe obesity confirms that health benefits proportional to weight reduction can be expected. It also suggests that even relatively limited levels of weight control could be beneficial. The observation of a smaller excess mortality in recently recruited individuals might be due to improved therapeutic strategies, including diagnosis and treatment of complications.
Data collection for this study was supported by a grant from the Regione Piemonte (Ricerca Sanitaria Finalizzata 2002/146). The work of D.M., M.C., and F.M. in study design and conduct was partially supported by grants from the Special Project Oncology, Compagnia di San Paolo, and from the Associazione Italiana per la Ricerca sul Cancro (Italian Association for Research on Cancer). We are indebted to Marina Marin for data collection and management at the Istituto Auxologico Italiano and to Paolo Avagnina, Sebastiano Avagnina, Franco Contaldo, Francesco Caviezel, Gabriella Garruti, Riccardo Giorgino, Fabrizio Pasanisi, Marco Tinivella, and Antonietta Tufano for data collection at the other centers involved.
The authors declared no conflict of interest.