Our results indicate that from 1995 to 2009 prescribing, broadly speaking, corresponded with availability, licensing and guidelines. Key findings are: 1) the proportion of patients offered treatment for bipolar disorder increased markedly between 1995 and 2009; 2) patients spent ever increasing amounts of time on psychotropic medication, in particular second generation antipsychotics and valproate; 3) this increase in time on medication was most noticeable in younger women; 4) antipsychotic and valproate prescribing increased relative to lithium; 5) use of second generation antipsychotics accelerated; 6) prescribing more than one drug at once increased; 7) treatment was not influenced by social deprivation and; 8) by 2009 one third of women of childbearing age who took medication for bipolar disorder were taking valproate. This final finding is worrying as guidelines are now very clear that valproate should be avoided because of its teratogenic potential.
The long-term management of bipolar disorder is complex. The prescribing recommendations described in the 2006 NICE guidelines 
should represent the gold standard, but they are unusually vague and recently a review and update has been requested 
. NICE recommends lithium, olanzapine or valproate as first line, which we have found to be the three most commonly prescribed psychotropic medications for maintenance. If the patient has frequent relapses, or symptoms continue to cause functional impairment, they recommend switching to an alternative monotherapy or adding a second prophylactic agent (lithium, olanzapine, valproate). However the evidence supporting the maintenance use of olanzapine and valproate is limited, and valproate is not licensed for this use. If anything, the evidence from research carried out after 2006 further strengthens the argument for use of lithium first line 
, but the complications of initiation, monitoring and side effect profile may continue to limit its use. It is also recognised that irregular use of lithium produces poor outcomes 
, risk of relapse on stopping 
and that less than two years use may have no beneficial effect 
. It may therefore be that the degree of concordance suggested by our findings reduces its benefit relative to other maintenance medications.
The increase in time spent on medication is likely to represent both increased prescribing and increased adherence to medication 
. Previous studies of antipsychotic prescribing trends have found that, over time, patients have been prescribed medications (for all indications) for longer periods 
, and this has been shown specifically in the bipolar disorder patient group 
. Our results differ from some studies from the United States, which found that lithium prescription for bipolar declined, over the period 1990–2005 
. The latter two studies also failed to show the increase in antipsychotic prescriptions found in this study.
It appears that use of second generation antipsychotics for bipolar disorder pre-empted the available scientific evidence. The first case reports suggesting the effective treatment of mania with olanzapine were published in 1997 
. These were followed by the publication of the first randomised control trials in 1999 
. Olanzapine was first given marketing authorisation for psychosis in 1996, and was first used in our cohort in 1997. Risperidone was first used in our study in 1996, two years before the first published randomised control trial into its effectiveness as monotherapy 
, but four years after it was first authorised. Quetiapine, authorised for schizophrenia initially in 1997, was first used in our sample in 1998, while the first trial of quetiapine as an add-on medication was not published until 2004 
. However it does seem reasonable that clinicians made inferences from trials of second generation antipsychotics in psychosis and clinical experience of first generation antipsychotics.
Concerns were raised about the teratogenic effects of valproate in the early 1980s 
, but it was only in 2004 that the risk was confirmed to be higher than other anticonvulsants 
. Therefore psychiatrists prescribing in this population may only have become aware of this risk via the NICE guidelines from 2006 onwards 
. However, despite this advice, use of valproate has continued to rise since then.
Previous studies have shown that co-prescribing is common 
with up to 80% of patients on a mood stabiliser plus another medication. In our cohort 48.1% were prescribed more than one agent in 2009, and an anticonvulsant plus an antipsychotic became the most commonly used dual therapy. Co-prescribing, although concordant with guidelines is problematic due to the side effect profiles of the drugs used, and concerns over long term health risks 
THIN is a primary care database and, as with all clinical databases, it is impossible to be sure that a person prescribed a psychotropic medication was concordant. However, it is fair to assume that repeat prescription of a particular drug implies medication collection, from which we may infer some degree of adherence. Gaps in treatment may be explained by hospital prescribing, such as during acute inpatient stays, and therefore we may be underestimating the duration of treatment. Also, there may be a number of patients who receive all their medication from secondary care, though these numbers are likely to be small, given the manner in which prescribing budgets are allocated in the UK. Secondary care prescribing may have been higher earlier in the study time frame, but analyzing the results excluding the first 5 years of data does not change the findings. There is no reason to suppose that any particular group would have been preferentially prescribed for in secondary rather than primary care. Although the focus of this study is primary care prescribing, specialist treatment is likely to be a major influence on GP prescribing, and so it is likely that these trends would be reflected in the total population with bipolar disorder.
We were unable to separate bipolar I disorder from bipolar II disorder in our cohort; however given that treatment guidelines are the same for both subtypes of the disorder (extrapolated from research in bipolar I), the trends in prescribing are likely to be very similar. Changes in diagnostic practice probably mean there was an increase in the number of bipolar II patients over the study period. From our current study we are unable to comment on whether a drug was prescribed initially as a monotherapy or, if not, in what order it may have been added to the treatment regimen. We also do not know about historical prescribing for the cohort so we are unable to state whether clinicians have adhered to the guidelines for first line drugs on initiation of treatment. 62% of the treated sample were women, although the overall incidence in men and women is thought to be equal, it is recognised that females have more acute episodes of illness 
, and therefore their information may be better recorded in the database. Our cohort was also relatively old at the start of follow-up (mean 44.5 years) compared to the observed age of onset of the disorder, which tends to be in the early twenties 
. It is likely therefore that some our sample is previously diagnosed cases that have entered the database late. It may be that trends in treatment of newly diagnosed individuals (incident cases) differ from our findings.
This study identified broad concordance with prescribing guidelines. Our findings suggest a number of important trends that should be noted by researchers and clinicians alike, the most striking being the overall increase in prescribing and time spent in treatment. A number of questions remain unanswered about the long-term management of bipolar disorder. Although there is unlikely to be one ideal treatment for all patients with bipolar disorder, as the illness is heterogeneous and subtypes appear to be associated with a preferential response to specific drugs 
, further studies with long follow-up times are necessary to clarify the benefits (and risks) of different psychotropic medications, especially antipsychotics. Despite this it would be useful to prescribers (both psychiatrists and GPs) if NICE guidelines were able to be more precise about recommendations, especially in the areas of first line treatments and treatments for women of childbearing age. Perhaps there also needs to be more education about these areas.