This study presents hospitalization rates in a cohort of obese, adult patients recruited in a centre specialized in the treatment of obesity and its complications, compared them to those in the general population, and estimated absolute costs of hospitalization. Hospitalization rates in this cohort were much higher than in the general population, with SHRs increasing according to the severity of obesity, as measured by BMI or WC. Approximately 70% of all hospitalizations in the cohort were attributable to obesity, and the SHRs for most obesity-related co-morbidities were significantly increased.
The results presented here on the excess hospitalization of cohort members with severe obesity compared to the general population are in line with the results of Keating et al. on outpatient services and the utilization of pharmaceutical therapies .
Overall, a comparatively larger effect of obesity on hospitalization rates among males compared to females was found, with the highest risk in the age groups 30–39 and 40–49 years. Among females, age-specific hospitalizations ratios showed less variation. The major difference between the sexes was in the age group 30–39 years; males had the highest and females the lowest SHRs because of different expected baseline risks, probably due to maternity-related hospital admissions in the general female population. In line with Luchsinger et al.  and Queensberry et al. , the SHR increase at older ages (>70 years) was less pronounced in both sexes, even though the absolute number of hospitalizations due to obesity remained large in females. The SHR patterns by sex and age were consistent with the corresponding patterns in standardized mortality ratios described by Mirabelli et al.  in a previous cohort study of patients with severe obesity.
SHRs were higher in cohort members with severe obesity and a WC above the median in both sexes, but differences by category of severity were larger among males. However, in this study cohort the presence of complications was a prerequisite for admission to the IAI for individuals with a BMI of 30–39 kg/m2, but not with a BMI of ≥40 kg/m2. Consequently, differences in morbidity between the two BMI groups were likely to have been attenuated. The capability of WC to identify patients with a worse prognosis at least as efficiently as BMI has already been described .
Obesity is known to be an independent risk factor for CVDs [29, 30]. Many studies found a positive association between obesity and CVD-related outcomes [11–13]. In this study cohort, both sexes presented a higher number of hospitalizations due to CVDs compared with the general population (SHRs approximately 3.5 in males and 2.5 in females). No trend was observed with severity of obesity, but in this comparison the underestimation of differences due to the selection criteria applied to the study cohort may have played an important role. Particularly high SHRs (9.1 in males, 4.8 in females) for heart failure were also observed, most likely due to the well-established strain induced by obesity on heart function .
Obesity is associated with a wide range of respiratory conditions, including chronic obstructive pulmonary disease (COPD), asthma, obstructive sleep apnoea, pulmonary embolism and aspiration pneumonia . Nevertheless, little is known about the burden of hospitalization due to obesity-related respiratory complications. Han et al.  found no significant difference between the number of hospital admissions due to pneumonia and COPD in obese and normal-weight patients. Kornum et al.  found an increased risk of pneumonia-related hospitalization among males, but the association disappeared after controlling for other chronic diseases. In this study cohort, hospitalizations for respiratory diseases were four times more frequent than in the general population. There was a clear, albeit possibly underestimated, trend in SHRs according to the severity of obesity in both sexes. COPD had the highest SHRs, while pneumonia and influenza were only slightly more frequent than in the general population.
SHRs for cancer, both overall and site-specific, were not increased in males. In females a slight excess for all cancers was observed, without any evidence of trends by BMI or WC, and a small excess for cancers of the colon-rectum specifically. The association between obesity and cancer incidence and mortality is well established [10, 33–35]. However, it is important to note that this study included relatively young patients (mean age 50.7) and a limited median follow-up time of 7.3 years. Therefore, at present this cohort cannot give information on cancer hospitalization rates. Similar findings were reported in the Italian mortality cohort study .
Obesity is considered to be one of the most important risk factors for knee osteoarthritis . A Norwegian cohort study  also showed a significant association between obesity and hand osteoarthritis. Weitoft et al.  reported a 50% higher risk of inpatient care for musculoskeletal diseases among obese patients than normal-weight individuals. In the present study cohort, the SHRs for diseases of the musculoskeletal system and connective tissue were 2.6 for males and 3.8 for females, i.e., considerably increased, even if no trend for the severity of obesity was apparent.
Finally, hospitalizations due to mental disorders were in excess in this study cohort compared with the general population, particularly among females (SHR 6.8). Several studies [38, 39] reported an increased risk of mood disorders in obese individuals of both sexes. The higher SHRs among females compared to males may be attributed to an increased susceptibility to some disorders, such as bipolar disorders and social phobia .
SHRs for obesity, diabetes and rehabilitation were not reported for several reasons. First, these diagnoses are strongly associated with severe obesity. Second, diabetes is considered by the NHS as an inappropriate cause of hospital admission. Conversely, severe obesity can justify diabetes as a principle diagnosis in the HDR database. Consequently, SHRs for diabetes in the study cohort versus the general population could be over-estimated due to a differential attitude in filling discharge records. Third, the admissions due to rehabilitation procedures are likely to be partially covered by the first admission to the IAI.
It is important to point out that relative measures of association, such as SHRs, should be read in combination with absolute numbers to fully understand the burden of disease induced by obesity. Indeed, if considered independently, these two measures may be misleading. The hospital admissions patterns described in the present report determined a per-person mean annual cost of hospitalization of more than 3,000 euros. Since all cohort members were enrolled during a first admission to the IAI in Piancavallo, they all had at least one hospital admission, and non-zero costs of hospitalization.
The main determinants of hospitalization costs were analysed to quantify their contribution to the cost accumulation. The economic burden due to hospital use was highly affected by the presence of co-morbidities. Malignant neoplasms, respiratory diseases, hypertensive diseases and diseases of the musculoskeletal system and connective tissue were predictive of higher annual costs of hospitalization in both sexes. CVDs were a relevant cost determinant only for males, with a 24% increase in annual costs of hospitalization, whereas costs increased by 16% with the presence of mental disorders among females only.
Diabetes was not a significant determinant of hospitalization costs, but diabetes per se was expected to be a driver mainly of drug costs and outpatient care. Hospitalization of cohort members with diabetes was usually due to age and diabetes-related complications, mainly CVD. In fact, among cohort members with diabetes, CVDs were found to be a significant predictor of higher hospitalization costs in both sexes.
In the study cohort, the severity of obesity, as described by BMI, did not affect hospitalization costs. This result is in contrast to Andreyeva et al. , who described an increase in healthcare expenditures among moderate to severely obese males and among severely to extremely obese females. Other previous studies [19, 22] on BMI and healthcare expenditures are not comparable with the observations in the present study, as they used the BMI of normal-weight patients (BMI 20–25 kg/m2) as a reference. The cohort selection criteria likely affected the association between BMI and costs in the present study by reducing the differences associated with BMI. WC was a significant predictor of hospitalization costs among females, with an 8% increase in costs for every 10 cm increase in WC. Previous studies suggested that abdominal adiposity may be a better predictor of healthcare costs than BMI [24, 41].
The main strength of this study was that it was based on a large cohort, with baseline information available on several well-recognized risk factors for subsequent morbidity and mortality, including smoking, alcohol consumption, and various clinical parameters, which allowed us to adjust for potential confounders. However, a number of limitations must be considered when interpreting the results of this study. First, the vital status of cohort members was not systematically ascertained. Cohort members who died during a hospital stay could be identified, but all others were assumed to be living throughout the observation period. As a consequence, overestimation of the person-years in the study cohort may have occurred, and led to an underestimation of SHRs, even if the low mean age of cohort members is likely to have kept any underestimation small. Second, the cohort was enrolled at a specialized centre, and therefore should not be assumed to be representative of the overall obese population in Italy. Nevertheless, as the IAI offers a residential nutritional rehabilitation programme, but not active treatment of obesity (such as bariatric surgery) or of complications due to obesity, the impact on subsequent hospitalization patterns due to complications is likely to be limited.
Reported hospitalization rates have been standardized by age, sex and calendar year. A potential confounder of hospital use could be socioeconomic status. Nevertheless, due to the lack of information in the general population dataset, it was not possible to adjust for socioeconomic status.
Finally, costs were calculated using DRG tariffs, which are used to reimburse healthcare providers. DRG tariffs do not represent actual costs, but standard costs. While obese people have been reported to experience longer hospital stays , and to be likely to require more resources compared with normal-weight individuals affected by the same diseases, the DRG tariffs did not vary based on the presence of obesity, and therefore real costs were probably underestimated.