I am in the process of analyzing mortality statistics for those who come through the public behavioral health system in Ohio (e.g., severely mentally ill - SMI). To this end, I will be calculating Standardized Mortality Ratios (SMR) for those who are diagnosed with various conditions compared to the general population. Many people pass through our system each year (>400k), but relatively few of these people die each year. I will have to pool our figures from 2004-2011 to generate stable SMRs that we can stratify into subgroups.
For decedents in the general population, the observed person years will be equivalent to one month per person for each month of the study until they die, starting on Jan 1, 2004 (e.g, Date of death – Jan 1, 2004 = total months observed which I will convert to years).
I can do a couple different things for decedent clients, but I’m not sure which is correct.
1) I could treat any decedent client as “exposed” to SMI and calculate the equation the same way I do for the general population.
2) I could calculate the number of months enrolled in treatment until a person dies (while being treated) and use that figure as the number of years observed.
3) I could calculate the number of months enrolled in treatment until a person dies (while being treated or after discharge from service) and use those figure as the number of years observed.
4) I could be more imprecise and assign a client a value of 1 year in the year that they died if they received treatment that year.
5) Something else?
It is not entirely clear to me which information you have in your database, but you should count the time they are 'exposed' (or, in other words: diagnosed with SMI) until they die... so the best measure you have for that. I am not sure if all SMI patients are treated continuously, or from diagnosis forward, or if treatment is sometimes delayed. If that is the case, ideally you should find a way when SMI initiated... unless your question is if the TREATMENT influenced mortality, then you can just take time exposed to treatment.
An excellent study. I am just curious, do you plan to adjust for age, gender, ethnicity/race, and possibly county or Zip Code (as a proxy for socioeconomic factors associate with access to care)? I think each of these factors can have an impact on mortality rates.
@Esther: I already have most of the death data from the Ohio Dept. of Health, which includes variables like age, race, gender, NCHS code, date of death, etc. These data will be linked with our client records to determine who passes through our system. Our databases contain numerous variables including admission date, last date of service, case closure date, previous episodes, etc. Treatment can be delayed, but I also have a variable that is date of first contact.
Exposure is a difficult target. I can tell you when we first saw them in the system, but that doesn’t mean the first visit represents the onset of disease. (They could have been seen out of state, out of country, in a private system, by a primary care doc, etc). I would love to know the disease onset, but I don’t. Perhaps I should only calculate SMR’s for clients where previous episodes =0 because the admission date is more likely to represent the onset? Or I could just look at treatment influenced mortality like you suggest. If I go this route, then I still have questions. If I assume that all SMI is lifelong, then I could use technique 3, but if I assume it’s recurrent episodic, then I have to use technique 2 and use their most recent episode in treatment. Both are possible, so I’m not really sure which one I would choose…
@James: I plan to adjust for age, gender, and race when possible. Adjusting for/stratifying across geographic areas (i.e., census tracts, zip codes, and counties) is possible, but some of the areas experience very little mortality, so the SMRs are typically considered unstable. I have better luck when I adjust for regions (e.g., urban vs. rural).
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology