Our main findings were: (I) a strong association between use of medication to treat diabetes mellitus and mood stabilizers for the treatment for bipolar disorder, and (II) females had a 30% higher risk compared to men of being treated simultaneously for both disorders. We also found (III) that persons using oral anti-diabetic agents had higher odds of receiving valproate than either lithium or lamotrigine. (IV) Persons receiving insulin in monotherapy seemed to have a lower odds ratio than persons receiving oral anti-diabetic agents of also receiving mood stabilizers, compared to the general population.
The present findings are in agreement with previous studies showing comorbidity between diabetes mellitus and bipolar disorder. Two studies have shown a prevalence of diabetes mellitus of 10% in a group of hospitalized bipolar patients [1, 2]. A study  found a 26% prevalence of type II diabetes among patients with bipolar 1 disorder. Another study reported  a 11.7% prevalence of diabetes in the bipolar population, and that patients with comorbid diabetes mellitus and bipolar disorder had a 81% disability rate, compared to 30% in the non-diabetic bipolar population. None of the studies noted whether this comorbidity differed by gender, or if the patients were affected by medication use. Carney  investigated medical comorbidity in women and men with bipolar disorder. They found diabetes with complications to occur at increased odds (adjusted OR = 1.54) among 3,557 subjects with bipolar disorder, compared to the non-bipolar patients in the sample group. However, the odds of having uncomplicated diabetes was not significantly elevated compared to the control population. This finding may be related to the fact that individuals with psychiatric disorders tend to get poorer and more sporadic medical treatment than the general population. Carney et al. found that more woman (61%) than men (39%) were identified with bipolar disorder. However, they did not investigate whether comorbidity of bipolar disorder and diabetes mellitus was different for males and females. Fiedorowicz  found a higher prevalence of diabetes (30%) in 217 bipolar patients, compared to the general population (12.5%). In a small sample (women n = 25, men = 18), they found men to have a higher prevalence of diabetes mellitus (33%) than women (28%), but found no significant association between medication use and cardiovascular risk factors.
Kessing et al.  investigated whether some of the co-morbidity between diabetes and bipolar disorder could be the result of conditional probability, or Berkson's bias. Patients with osteoarthritis were compared to patients with bipolar disorder. They found a slightly increased rate of diabetes among people with bipolar disorder compared to those with osteoarthritis (aged between 45 and 80 years). However, the prevalence of diabetes mellitus may have been underestimated and the control group of osteoarthritis is not ideal , due to a not yet understood association between osteoarthritis and insulin [14–16]. For example, osteoarthritis and diabetes share an etiological role of inflammatory pathways, as do mood disorders .
Three of the four mood stabilizers analyzed in the present study (carbamazepine, valproate and lithium) all have the potential to increase weight and, therefore, may increase the risk of diabetes type II. Mood stabilizers may also increase risk of diabetes mellitus in other ways than weight gain. Insulin tends to be used more for type I diabetes, and oral hypoglycemic agents more for type II diabetes, and the linkage between diabetes and bipolar disorder is stronger for type II than for type I diabetes. Valproate and lithium both inhibit glycogen synthase kinase 3b [18, 19], as does insulin.
The finding of the high diabetes prevalence in patients with bipolar disorder could indicate that those with this disorder are at higher risk for diabetes by virtue of a shared pathophysiology common to both disorders, such as inflammation [20, 21]. There is a genetic risk factor in both diabetes mellitus and bipolar disorder [22, 23].
Women tend to experience more side effects from bipolar treatment , particularly weight gain, which may be a factor explaining the increased OR we found in women compared to men.
Strengths and limitations
The results are based on data from the total Norwegian population, and the risk of selection bias is low. Our sample was initially large; however, only 900 persons received medication for both bipolar disorder and diabetes mellitus. This number is nevertheless higher than in any other study we know. The lack of inclusion of data from hospitalized and institutionalized patients probably represents a small number, as few patients stay in the hospital throughout an entire year and the majority of the institutionalized patients are 70 years and older in nursing homes. Mood stabilizers are used less frequently by persons younger than 20 and were, therefore, excluded. Treatment with insulin is not indicated for any other disease than diabetes mellitus. Persons at risk for developing diabetes mellitus type II rarely use oral anti-diabetic agents. In Norway, metformin is rarely used prophylactically in psychiatric patients. Lithium is to some extent also used for treatment of unipolar depressed patients; excluding persons on antidepressants did not substantially change the OR.
Mood stabilizers might be used by patients diagnosed with other psychiatric diagnoses, such as schizophrenia, and this presents a possible confounder, as the prevalence of diabetes is increased in patients with schizophrenia . We are not aware of data from Norway on the use of mood stabilizers in schizophrenia, and data from other countries vary widely [26, 27] so it is difficult to know how important this confounder may be. Of probable greater importance is the ability of antipsychotics to induce diabetes . However, when adjusting for the use of antipsychotics the odds of receiving medication for bipolar disorder and diabetes mellitus was still significantly elevated (1.3). When excluding all persons using any antipsychotic agents, the odds increased to 1.6. Antipsychotic medication can possibly account for some of the correlations we found, but not all. Ruzikova  found that patients with bipolar disorder had a greater chance of complicated diabetes mellitus. It is therefore possible that patients with severe bipolar disorder, receiving both mood stabilizers and antipsychotic treatment, may have a higher risk of also receiving medication for diabetes mellitus. This needs to be investigated further.
The prescription database indirectly measures the frequency of diabetes and bipolar disorder in the adult population, and our risk estimates may, therefore, be interpreted as a proxy of comorbidity between the diseases. The validity of this measure is likely high, as all registered prescriptions are based on a physician's examination, diagnostic decision, as well as the fact that the patients have collected their medication. The study does not include persons who either do not seek treatment, recognize their illness or for any other reason do not collect their medication. As such, the study may have under-estimated the true prevalence and, thus, comorbidity. The cost for medical treatment is to a large extent covered by the public health care system in Norway, and very few persons will fail to purchase medication for economic reasons.