I was all set to continue my crusade against the extremely high prevalence estimates for mental illness when I started to notice some discrepancies in the data. So that plan is on the back burner for a little while until I run through the numbers.
The biggest issues I have noticed is that estimates for the prevalence of mental illness in my home country of Canada are not always consistent with those in the US. So here’s a little background on the story.
Starting in the 1980s, a number of large-scale epidemiological surveys were conducted, mostly by sending interviewers into communities. The first major study of this sort was the Epidemiological Catchment Area (ECA) survey conducted at mostly urban sites in the US between 1980-1985 and involved around 20,000 participants (~2000 were inpatients, the rest lived at home). This survey used a questionnaire called the “Diagnostic Interview Schedule” (DIS), a proxy for the diagnosis of a DSM disorder outside of the doctor’s clinic. The ECA study was “prospective” which means that the authors contacted individuals twice, one year apart, and estimated whether they had experienced an illness at some point during this period. This first study used the criteria from DSM-III, which did not require that symptoms have a significant impact on a person’s life.
I have graphed the prevalence estimates from this study and a number of others that I will discuss below, both for lifetime estimates (the proportion of people who will have a mental illness at least once in their lives) and 12-month estimates, based upon individuals who reported symptoms consistent with a DSM disorder within the past year.
Around the same time (1983-1985) another metropolitan study of about 3,200 people was being conducted in communities within Edmonton, Alberta. This study also interviewed people using the DIS and made diagnoses according to the DSM-III.
A second run of studies with a much broader scope were undertaken in the 1990s. One of these study was the National Comorbidity Study (NCS), conducted between 1990-1992, which recruited a national sample of 8,100 individuals from the US, aged 15-54. This study used a state-of-the-art version of the DIS called the Composite International Diagnostic Interview (CIDI) which allows non-clinicans to make highly reliable diagnosis of mental illness. This study diagnosed people using a newer version of the DSM-III-R, and recorded both 12-month prevalence and lifetime prevalence. The Ontario Health Survey (available here), used similar methods to study 10,000 individuals, aged 15-64, living in Ontario communities in the fall of 1990. These studies were supplemented by the NCS-Replication Study (data available here) which was conducted between 2000-2002.
What is striking about these studies is just how high the mental illness prevalence estimates are. Take a look at Figures 1 and 2 (right click to “view image”):
Each table indicates the prevalence of individual “mood”, “anxiety”, “substance dependence”, “impulse control” and “other” disorders. For the first four groups of disorders, I have included a “total” for each mental illness category, which is the rightmost data point of each cluster. Lifetime prevalence (Figure 1) across the Edmonton study and the two NCS studies is 10-35% for each of the mood, anxiety and substance dependence clusters, and total lifetime prevalence is from 20% (excluding substance disorders) to 55% of the population. Yearly prevalence estimates (Figure 2) are in the range of 20-35% when we include substance dependence and still account for 15-25% when substance abuse is ignored.
An interesting point to note in the 12-month prevalence data is that the Canadian studies, even though they used quite similar methodology to the American ones, indicate markedly lower prevalence rates on the order of 5-10%.
Nonetheless, these numbers are striking for a few reasons. For one, they indicate that in any given year, between 1 in 5 and 1 in 3 North Americans might be suffering from some sort of mental illness. This is, frankly, implausible by any reasonable definition. Another point is that it doesn’t match what appears to be written all over the sites of mental health groups in Canada. Many cite a 1-in-5 figure for lifetime prevalence and those which list sources link back to a a 2001 Report on Mental Illness in Canada, which cites the Edmonton and Ontario studies I discussed above and/or the 2002 Canadian Community Health Survey (CCHS): Mental Health and Well-being Survey. However, as far as I can tell, there are no published lifetime estimates from this survey (although I am currently trying to get a hold of the source data).
So are mental illness rates really this high?
Well, those who cite the CCHS study clearly might think not, as this study indicates a 10.6% 12-month prevalence of mental illness in Canada (still quite high), which is half that of most other studies. However, I have also graphed the available data from this study (Figure 3), which includes only 2 mood disorders and 3 anxiety disorders (ignoring, for example, simple/specific phobia which tends to account for 3-5% of cases and dysthymia which often accounts for 1.5-5% of cases; but also post-traumatic stress disorder, generalized anxiety disorder and obsessive compulsive disorder) and has a very strict definition of substance dependence (other studies include “substance abuse”)–not to mention no inclusion of other when-known disorders like ADHD, schizophrenia, etc. So really it is only when we exclude nearly half the diagnoses contained within the DSM that yearly prevalence gets even close to 10%.
For comparison, I’ve graphed the data in the third figure against a re-estimate of the ECA and NCS data conducted in 2002, which requires that individuals meet criteria for clinically significant and/or functional impairment. Clinical significance here is defined as another who has the symptoms of a disorder who has also done one of the following: i) talked to a health care professional about symptoms ii) found that symptoms interfered with everyday life or iii) is taking medication for symptoms. I’ve also graphed what the authors of this study propose as revised estimates for individual disorders (“Revised US”). This basically just looks like they picked the lowest estimate from either the ECA or NCS data an then used that lower count as an official prevalence estimate.
These revised criteria cut down significantly on the 12-month prevalence of mental illness, but they still greatly outstrip the CCHS data. In other words, the most comprehensive, nation-wide survey of Canadian mental illness appears to greatly underestimate mental illness rates, mostly because it just doesn’t count more than a handful of mental illnesses.
A point to make about these revised criteria is that although they appear to reduce the prevalence of mental illness into a more reasonable range (i.e. which is less likely to categorize people with “typical” life problems as being mentally ill), it’s still not clear that this study identifies the people who actually have mental illness. That’s because what’s actually happening is these new criteria re-define illness in terms of treatment need or treatment-seeking. But in many cases individuals with true mental illness may not seek treatment, and those without may very well go to the doctor for medication anyways. So this doesn’t answer the most important question for us: how many of these people (as many as 15-20% per year, see Figure 3 above) who have the symptoms of mental illness and who are also seeking treatment may not actually need that treatment? And how many individuals with real disorders, who either don’t have severe symptoms yet but soon will, or whose potentially severe symptoms don’t meet the cut (e.g. about 30% of people with suicidal ideation, significant loss of work were excluded according to the new criteria), aren’t getting the help they need?
I will try to begin addressing the first group in a future post with some data on prescription drugs use for mental illness.