I am currently writing my first paper and therefore relatively new to the Operation Research community. I have a physician scheduling model with novel demands and now I want to test it. Since the model is rather simple and should provide new insights and not concrete shift schedules, I don't think it's worth using real data as demand patterns. Instead, I want to construct artificial demand patterns. Is there a general convention on how to generate such patterns? What are the usual criteria for this? What constitutes a suitable pattern? And is it usual to first create a demand pattern for the analysis part, then evaluate it and then check in the sensitivity analysis whether the collected results are robust to these changes, or are there other approaches?
3
-
$\begingroup$ This will depend on the nature of the physician's employment. Demand patterns for an ER doctor will, I suspect, be more random than demands for a specialist who sees most if not all patients via appointments made well in advance. $\endgroup$– prubin ♦Commented May 29 at 20:20
-
$\begingroup$ @prubin Thanks. Well i assume the ICU as the part of interest, where the daily demand is rather constant. In this case, would a Poisson approach be suitable? $\endgroup$– ornewbieCommented May 31 at 9:39
-
$\begingroup$ Treating arrivals as a Poisson process sounds plausible, although I would recommend looking at historical data to confirm that. If you are talking about exponential service times, I'm not so sure. I'd like to think that the longer a patient is in the ICU, the closer they are to getting out. $\endgroup$– prubin ♦Commented May 31 at 15:53
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
You should checkout this question. The answers state that often Poisson distributions are used due to nice properties and as they are often quite good representations of actual demand arrival patterns.
You can also vary your demand based on the time of the day or the week day. If you would use a Poisson distribution, you can for example set a different rate for each hour of the day and then generate different patterns for various days.
However, I want to stress out that generating demand patterns is highly dependent on your problem. You could speak to a physician and ask how demand usually arrives in the system. It could be that many patients arrive early in the morning before the hospital/practice opens.
-
$\begingroup$ Thanks for your answer. Would a Poisson distribution still work if I do not have time intervals per se, but rather q fixed number of shifts per day? And if I also assume a rather constant daily demand, e.g. in the ICU, where the daily demand is rather constant, independent of weekdays or weekends? At what point in the day does the demand change? $\endgroup$– ornewbieCommented May 31 at 9:38
-
$\begingroup$ The demand distribution should not depend on the shifts, but should be seen as an exogenous parameter. If you assume constant demand then you can just use one parameter (rate). You can also look at papers that considered ICU planning such as this one and apply their parameters. It would be valuable to see how your model reacts to higher than usual rates, sth I assume is very important in ICU planning. $\endgroup$– PeterDCommented May 31 at 9:49