Detecting changes in incidence rate

  • Context: Graduate 
  • Thread starter Thread starter Gerenuk
  • Start date Start date
  • Tags Tags
    Incidence Rate
Click For Summary

Discussion Overview

The discussion centers around detecting changes in the incidence rate of rare events over time, specifically focusing on identifying points at which the frequency of these events increases. Participants explore statistical tools and methodologies for analyzing such changes, considering various temporal profiles and the implications of random variation versus real trends.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested
  • Mathematical reasoning

Main Points Raised

  • One participant seeks a solid statistical tool to analyze the increase in event rates and to determine if observed changes are statistically significant or merely random.
  • Another participant suggests using moving averages as a robust method adaptable to different temporal profiles, along with linear regression models for sustained trends.
  • A reference to a paper on Autoregressive Integrated Moving Average (ARIMA) is provided as an example of a combined approach for analyzing trends in incidence rates.
  • Concerns are raised about the adequacy of regression analysis in providing confidence intervals for trends, particularly given the limited number of events (about 200) in the dataset.
  • A suggestion is made to consider control charts as another potential analytical tool for monitoring changes in incidence rates.

Areas of Agreement / Disagreement

Participants express varying opinions on the best statistical methods to use, with no consensus reached on a single approach. The discussion reflects differing views on the effectiveness of regression analysis and the importance of controlling for confounding factors.

Contextual Notes

Participants note limitations related to the small sample size of events and the potential impact of confounding factors on the analysis. There is also mention of the need to minimize beta error when assessing trends.

Gerenuk
Messages
1,027
Reaction score
5
Hi!
I have rare events over time (about 200 incidences) and I'd like to detect when the time rate of these increase, i.e. at which point more events per time start to occur.
Do you know a solid statistical tool to analyze that? I'd also like to know if the change is real evidence or just random.
 
Physics news on Phys.org
Gerenuk said:
Hi!
I have rare events over time (about 200 incidences) and I'd like to detect when the time rate of these increase, i.e. at which point more events per time start to occur.
Do you know a solid statistical tool to analyze that? I'd also like to know if the change is real evidence or just random.

It's somewhat dependent on the temporal profile: isolated random spikes, periodic spikes or waves, or sustained trends. Generally, various applications of moving averages are often used. These are fairly robust and can be adapted to different profiles. You can look up "Moving Average" on the Wiki for a fairly good general discussion. Linear regression models also work well for more sustained trends where you want to test for trends greater or less than 0.

The following paper shows an application of a combined approach called Autoregressive Integrated Moving Average (ARIMA) for dengue fever.

http://www.ajtmh.org/content/79/6/933.full.pdf

EDIT: As an aside, sometimes just inspection is more appropriate than statistical testing if public safety is an issue. You really want to minimize beta error (missing a trend that actually exists). You also will need to control for confounding factors. Differences in reporting and/or detection over time is often the case.
 
Last edited:
  • Like
Likes   Reactions: WWGD
Thanks a lot! I hope I can make use of the information.

My data is most likely something like a constant rate up to some time and then an increase in frequency. The problem is that it is only about 200 events in total.

For regression I'm afraid it won't tell me if the trend is with the confidence interval. But I need to read more about it...
 

Similar threads

  • · Replies 6 ·
Replies
6
Views
3K
Replies
1
Views
2K
  • · Replies 8 ·
Replies
8
Views
5K
  • · Replies 7 ·
Replies
7
Views
2K
  • · Replies 12 ·
Replies
12
Views
5K
  • · Replies 2 ·
Replies
2
Views
4K
  • · Replies 0 ·
Replies
0
Views
362
  • · Replies 16 ·
Replies
16
Views
3K
  • · Replies 12 ·
Replies
12
Views
2K
  • · Replies 10 ·
Replies
10
Views
4K