Psychosocial and Socioeconomic Risk Factors for Premature Death in Young People With Type 1 Diabetes

The assembly of the Diabetes UK (formerly British Diabetic Association) cohort of patients with insulin-treated diabetes from diabetes registers compiled throughout the U.K. has been described elsewhere (1). The earliest subjects in the cohort entered follow-up in 1972 and the latest in 1993. In all, the cohort includes >23,000 subjects with insulin-treated diabetes diagnosed under the age of 30 years. Details to identify each patient were sent to the relevant National Health Service Central Registers, and we were subsequently notified by them of all deaths and emigrations in cohort members.

This present study was restricted to southeast England and comprised the registers from Leicester, Poole, and the southeast component of the Diabetes UK register. These registers were chosen for logistical reasons because the research nurses who were to extract the data from hospital case notes were based in London. Where possible, two control subjects per case were selected. These were selected from within the same register as the case, and also matched on sex, date of birth (within 6 months), and date of diagnosis of insulin-treated diabetes (within 1 year). If the cohort contained no matched control subjects within the above criteria, then the date of birth and date of diagnosis criteria were extended by 6 months.

For each case and control subject, we attempted to find the hospital case notes for the study period, and if a subject had moved hospitals during this period, we contacted all of the relevant hospitals. Data were extracted by systematically searching through the case notes for the period from the date of diagnosis up to the date of death for the cases and from the date of diagnosis to the index date (the date of death for the matched case) for the control subjects. There are arguments to be made for using age, duration of diabetes, or date of death when defining the index date, but given the matching system, there is little practical difference. We extracted from the case notes details of the initial illness and diagnosis, clinical details including any complications of diabetes, and past medical history and family history. We also collected data on a number of psychosocial and socioeconomic variables measured at the index date such as unemployment, marital status, and whether the subject was living alone. In addition, we recorded whether a history of drug or alcohol abuse, smoking, or psychiatric referral had ever been reported. From the postcode at the index date, we obtained the ACORN (A Classification of Residential Neighborhoods) score (12), a measure of social profile that categorizes the subject’s underlying lifestyle and demographic characteristics as indicated by their postal address. For those variables that are not routinely recorded in hospital notes and might only be recorded if the variable was positive (e.g., a history of alcohol abuse), we used a category of “no mention” rather than “not known,” which we applied if we could find no reference to the variable in the case notes and interpreted this as no history of that particular variable. For those variables, “no mention” was taken as the baseline in the analysis.

The cause of death was taken as the underlying cause of death reported on the death certificate. This had been independently coded according to national criteria and was independent of the data abstraction conducted for the study. Since deaths coded to the underlying cause “diabetes” might reflect a range of different short-term metabolic events or long-term degenerative complications, the full statements relating to the cause of death were examined and all cardiovascular and renal deaths were recoded to the relevant categories. The cases were grouped according to whether the death was considered to be an acute event, a chronic consequence of diabetes, or unrelated to diabetes. The acute death category included sudden episodes of diabetes such as hypoglycemia or coma, as well as deaths from accidents and violence. All deaths from causes related to renal or cardiovascular disease that were plausibly diabetes related were included in the chronic death category.

Risks of death in relation to the study variables were estimated by calculating odds ratios (ORs) and CIs using conditional logistic regression methods for matched case-control studies (13). The difference between the ORs for acute and chronic causes of death were evaluated by fitting a model with an interaction and carrying out a likelihood ratio test. The P values cited are two sided.

In total, there were 146 deaths of cohort members from the study registers who met the case criteria, and of these, it was possible to obtain case notes for 98. The majority of the 48 case notes that could not be obtained had been destroyed, either because the hospital had been relocated or because the death had occurred more than 10 years previously. For the rest, every effort was made to retrieve notes from the various hospitals that the subjects had attended. For each case for whom case notes were located, two matched control subjects were sought. For 39 cases, 2 control subjects could be found, but for 59, only 1 control subject was available, either because only one subject on the study file met the matching criteria or the notes for the other potential control subject could not be found. In total, therefore, 139 control subjects were included.

Among the cases there were 51 deaths from acute causes, which included 24 acute complications of diabetes but also deaths from accidents (15 deaths) and suicides or possible suicides (12 deaths). Chronic conditions related to diabetes accounted for 34 of the deaths (22 from cardiovascular disease and 12 from renal disease), and the remaining 13 deaths were not considered to have any association with diabetes and included deaths from cancer, gastric complications, and respiratory diseases. None of the deaths were unexplained (14).

Descriptive characteristics of the study sample are shown in Table 1. The majority of the cases (86.8%) and control subjects (90.5%) diagnosed were under the age of 15, and 87.7% of the deaths occurred between the ages of 20 and 39 years. Although there appears to be a small difference at older ages between the case and control subjects in terms of age at diagnosis, this is because there are more often two rather than one control subject for the cases diagnosed at a younger age; therefore, this does not bias the analysis.

The ORs for mortality from all causes in relation to psychosocial and economic variables are shown in Table 2. Living alone, unemployment, a high ACORN score (indicative of poor social profile) at the index date, and a past history of drug abuse and psychiatric referral were all related to all-cause mortality. There was no relation between a history of either smoking or alcohol abuse and all-cause mortality. The ORs are also shown separately for the group dying from acute events and the group dying from chronic conditions. Deaths from acute events, but not from chronic conditions, were significantly associated with living alone, drug abuse, and past psychiatric referral. The OR for acute deaths was significantly higher than that for chronic deaths for psychiatric referral (P = 0.029). A high ACORN score and unemployment were significant risk factors for dying from chronic conditions. Although it was not possible, for statistical reasons, to calculate a matched OR for unemployment because the numbers of unemployed control subjects were so small (only 1 of 47 control subjects were unemployed at the index date compared with 12 of the 34 cases), the risk was clearly highly significant and the OR was significantly higher for chronic deaths (P = 0.002).

The risk of death in relation to known complications of diabetes before the index date is shown in Table 3. Death from all causes was significantly associated with hypertension, renal disease, and diabetic neuropathy, although not significantly associated with retinopathy. However, when the deaths were analyzed by cause, there was no evidence of any association between these complications of diabetes and deaths from acute events, whereas there was an association between renal disease, retinopathy, neuropathy, and hypertension and deaths from chronic conditions. The ORs for chronic deaths were significantly higher than those for acute deaths for hypertension (P = 0.041), retinopathy (P < 0.001), and renal disease (P = 0.011).

Previous results from the Diabetes UK cohort have shown that from the age of 30 years onwards, cardiovascular disease is the predominant cause of death in patients with type 1 diabetes and that deaths from renal disease are also common in this age-group (2). However, acute diabetes-related events, such as ketoacidosis, were responsible for the greatest number of deaths in early adulthood, closely followed by deaths from accidents and violence. In both men and women with type 1 diabetes, the mortality rates for death from these causes were much higher than in the general population. Within the Diabetes UK cohort, the young men were more vulnerable than the young women and the rates of death from acute complications of diabetes in men aged 20–39 were twice as high as the rates in women. The rates of death from accidents and violence were almost three times as high.

The present study was conducted to investigate the hypothesis that risk factors for death from acute events in young patients with type 1 diabetes may be different from those associated with death from long-term complications of the condition and that psychosocial and socioeconomic risk factors may be more closely related to death from acute events. Deaths from acute diabetes-related events were grouped together with deaths from accidents and suicides, as these two causes together accounted for all the unexpected deaths in this age-group. It is possible that there is a degree of overlap between these two categories; for example, an accident may have resulted from a hypoglycemic attack and a possible suicide from an inadvertent insulin overdose.

Inevitably, a case-control study of this type is dependent on the quality of the case notes, but for a mortality study, these may be the only available source of data. Despite great efforts to gather data as completely as possible, a number of case notes could not be found, in many instances because the notes had already been destroyed. It seems unlikely, however, that this would have resulted in bias because note destruction is related to the general notes policies of hospitals and not the characteristics of specific cases. There was also potential for bias as a result of missing data within the case notes that could be examined. In general, however, the proportion of subjects with such missing data was low enough, in relation to the size of the relative risks found, that bias of this type seems unlikely to explain the results.

To our knowledge, no other studies have examined risks of death subdivided by cause, and there have been few previous case-control studies of factors related to all-cause mortality in young patients with diabetes with which to compare our findings. Results from a case-control study from Pittsburgh in 1985 (6) indicated that the presence of renal complications was most strongly associated with premature mortality together with, in men, early familial mortality, alcohol consumption, and lack of participation in school sports. More recently, in a study similar to ours, a case-control study of 90 matched pairs was reported from the Japanese arm of the Diabetes Epidemiology Research International cohort (8). Survival, after adjusting for renal disease, was related to better education and better compliance with the treatment regimen.

The association between psychosocial factors and all-cause mortality in patients with type 1 diabetes has also previously been noticed from a number of cohort studies, although again the analyses have not been grouped according to cause of death. Smoking, low social class, leaving school before 16 years of age, living in a council house, and unemployment have all been shown to be associated with death after adjusting for renal complications (9–11).

In the present study, psychosocial factors were much more closely associated with deaths due to acute events than chronic complications. Living alone, a past history of drug abuse, and previous psychiatric referral were shown to be related to deaths from acute events and have not previously been identified as risk factors for all-cause mortality. The ACORN score, as a marker of socioeconomic status, was significantly related to all-cause death although not significantly related to acute event deaths once deaths were subdivided. Unemployment was also related to all-cause mortality but not associated with acute death, and although it was significantly associated with death from chronic conditions, it seems possible that unemployment was a consequence rather than a precursor of long-term complications of diabetes.

Other studies have shown that social factors are related to diabetic control in young adults (15) and hence give one mechanism by which acute deaths may occur. A supportive family background and better quality of life have each been related to better glycemic control (5,16,17), and a history of psychiatric disorders and not living in a two-parent home were shown to be related to noncompliance with medical treatment (4). However, there is no evidence that young people with diabetes in general have more social problems than those without diabetes (18,19). Instead, it would appear that the combination of diabetes and social problems leads to particular difficulties with compliance and poor control, which in turn might lead to acute death.

In this study, all-cause mortality was shown to be associated with hypertension, neuropathy, and renal disease, but it was interesting to note that none of these were associated with acute causes of death, only as the precursors of premature deaths from cardiovascular and renal disease.

Because the case and control subjects were matched for sex and age in this study, the analysis could not examine these variables as risk factors for mortality, although we have shown in an earlier report from the Diabetes UK cohort (see above) that young men with type 1 diabetes were particularly at risk of death from both acute complications of diabetes and accidents and violence. The present study adds several further factors with which to identify patients with diabetes at high risk of sudden death, and a combination of these findings and previous reports would suggest that these are as follows: young and of the male sex, a low socioeconomic background, living alone, a history of drug abuse, and past psychiatric problems. These groups are particularly vulnerable and would merit specific research and clinical attention.

The study was funded by Diabetes UK.

We thank the many individuals who worked to assemble the registers from which the cohort was identified, listed in Laing et al. (1). We recognize with thanks the roles of Drs. A. Bloom, D.R. Gamble, and T.M. Hayes in compilation of the original British Diabetic Association register and the input of physicians throughout the U.K. who contributed data to that register. We thank Dr. J.L. Botha, who contributed to the assembly of the Leicester register, and Dr. R.D. Hill, who contributed to the assembly of the register from Poole. We also thank the research nurses Maureen Swanwick and Pip Murnaghan and the computer programmer Zongkai Qiao.