# Thread: Temporal Distribution of Death

1. ## Temporal Distribution of Death

Dear all,

I plan to conduct a study. The preliminary aim is to examine the temporal distribution of death of terminal cancer patients in hospice ward (Monday to Sunday, and time of a day), to see if a particular day and time has more death and whether it associated with important Holidays, such as Christmas.

I have completed my scoping searches and have identified several relevant studies in this area (please see below). Some studies analysed the problem by comparing the "expected number of death" and "observed number of death" in days of the year, and to see if there is statistically significant difference; and in one study, "expected death counts were determined...using a quadratic polynomial regression model...which adjusted for seasonal variation in deaths from different causes". The authors have not elaborated much on the methods.

Can anyone provide expert opinion and some guidance as to the valid methods for my study. This will help me to ensure that I am using a correct statistical method.

Thank you very much.

Yours sincerely,
Michael Lee

1. Neumann CM, Hanson J, Kuehn N, Bruera E. Temporal distribution of deaths in cancer patients admitted to a palliative care unit. J Palliat Care. 1999 Autumn;15(3):10-3.
2. Panesar NS, Goggins W. Postponement of death around Chinese holidays: a Hong Kong perspective. Singapore Med J. 2009 Oct;50(10):990-6.

2. ## Re: Temporal Distribution of Death

Hi, since you want to relate your death probability to something happening within the hospice (I guess), it makes sense to correct for effects outside the hospice. Thus, it would be probably the best to have data from people dying outside the hospice, and than you can evaluate the difference. In the best case, you have data from people dying in the near / same city of your evaluated hospice, since probably the expected numver of deaths my vary from region to region (e.g. depending on the local climate).

Technically, you can probably use a Binomial regression model where the number of deaths in the hospice is the number of "success" and the number of deaths in hospice+number of deaths in the control group is the overall "number of trials". Thus, only if the proportion of deaths in the hospice compared to the number of deaths in the controlgroup changes, it is detected within the model. In this model, you can use "day" and "daytime" as predictors.

3. ## Re: Temporal Distribution of Death

Can you describe your potential data source(s)? How many patients and deaths with you have and for what time period?

4. ## Re: Temporal Distribution of Death

I have ~700 patient death in previous 1 year, each month roughly 50-60 deaths.
Can we detect whether a particular day or time of a day is having higher than usual death? Is it like describing or detecting a rise or fall in stock market?
But I don't know how to

5. ## Re: Temporal Distribution of Death

If you already have a function which gives you the "usual number of deaths" at a certain day and daytime (e.g. from other publication), an alternative to the above mentioned Binomial regression model is an offset-Poisson model, where "usual number of deaths" is taken as the offset. (If you use a Log-link function within the Poisson model which is the usual case, you use log(usual number deaths) as the offset). Like this, the regression model evaluates number of deaths in your hospice relative to the usual number of deaths. The advantage of this approach compared to the binomial model is that "usual number of deaths" can be non-integer (which is probably the case if it is some estimated smooth function)

6. ## Re: Temporal Distribution of Death

I am a newbie in medical statistics. But I guess hospice cancer patients have different patterns of death, compared with non-cancer patient (from other publication or source).
My data source is from a record book, charting down the death date and time, weekdays (Mon to Sun) in my hospice ward.
Is it possible to calculate the expected no. of death of each day (365 days per year) and each hour of a day (total 24 hours) respectively? Then compare the observed no. of death with the expected no. of death?

7. ## Re: Temporal Distribution of Death

hi,
as a first step, I would look at the autocorrelation of your time series. If there were any kind of periodicity - that should be visible there

regards

8. ## Re: Temporal Distribution of Death

Originally Posted by rogojel
hi,
as a first step, I would look at the autocorrelation of your time series. If there were any kind of periodicity - that should be visible there

regards
Thanks a lot Rogojel and all other experts.

Can I do the followings:
To list out the no. of death in each month, e.g. Jan 30, Feb 45, March 29, etc.
The expected death/month= total death in a year/12.
Then use chi squared test to see whether the distribution of death changes according to month of the year.

To list out the no. of death in each weekday, e.g. monday 100, tuesday 89, Wednesday 106, etc.
The expected death per each weekday = total death in a year/7.
Then use chi squared test to see whether the distribution of death changes according to weekdays.

To list out the no. of death in each hour of the day, e.g. 1am 100, 2am 89, 3am 106, etc.
The expected death per hour = total death in a year/24.
Then use chi squared test to see whether the distribution of death changes according to hours of the day.

9. ## Re: Temporal Distribution of Death

hi,
just a quick question, do you have only the number of death or also the number of patients? I mean the raw number of death will be impacted by the total number of patients, so you would definitely need to correct for it, right?

regards

10. ## Re: Temporal Distribution of Death

Hello Rogojel,
I have only the number of death in the hospice. But the ward has a quite stable bed occupancy throughout the year. Thank you.

11. ## Re: Temporal Distribution of Death

hi,
I would try to build some poisson regression models .

regards

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts