A method to detect a sudden increase in prevelance after introduction of outpatient registry

Objective

To detect, for each disease endpoint, whether there is a significant increase in prevalence in 1998 likely due to introduction of outpatient register

• Because many disease outcomes are missed before 1998:

  • Analyses that consider age at onset are not trustable
  • Bias is introduced due to swift change in prevalence

• Dataset:

  • Nationwide Finnish health registry data (~ 5.7 Million individuals)
  • Construct the same definitions of the FinnGen endpoints

We use Poisson Regression to model changes in disease prevalence

For each disease,

• Covariates:
  • event year
  • age
  • “discontinuation indicator”: 1 if the event year falls into 1999-2001; 0 if the event year falls into 1995-1997
• Output: prevalence of the disease
• Time frame: 1986 - 2011

Experimental setting to test model performances

Our model accurately identify endpoints with significant prevalence increase in 1998

Percentage of endpoints having a significant prevalence increase in 1998 is 70.6%

Percentage of endpoints with a significant prevalence increase in Neoplasms - �benign neoplasms vs malignant neoplasms: 60.1% vs 2.4%. Reassuring results because Malignant neoplasm are well covered by the cancer register since the 70’s.

A warning when a substantial prevalence change in 1998 is detected has been added to Risteys.

• https://risteys.finngen.fi
• An example: