prev next front |1 |2 |3 |4 |5 |6 |7 |8 |9 |10 |11 |12 |13 |14 |15 |16 |17 |18 |19 |20 |21 |22 |23 |24 |25 |review

Please refer to the MCHP report for the full list of case definitions investigated.  For this presentation, nine diabetes case definitions will be discussed.

 

Note: H = Hospital; P = Physician; Rx = Prescription Drug; PPV = Positive Predictive Value; NPV = Negative Predictive Value

 

The first case definition (a: 1+H or 1+P) is based on 1 fiscal year of data. Individuals are identified as having diabetes if they meet either of two conditions: if in a one-year period (here April 1, 2002-March 31, 2003) the individual has: 1) at least one hospitalization with diagnosis of diabetes (i.e., 1+H); or 2) at least one physician visit with a diagnosis of diabetes (i.e., 1+P).  Validation of this case definition resulted in Kappa=0.77, sensitivity =76.9% and specificity=98.7%.  PPV and NPV were 79.2% and 98.5%, respectively.

 

The second case definition (b: 1+H or 2+P) is similar to the first, with the exception that individuals are identified as diabetes cases if they have two or more physician visits (i.e., 2+P) with a diagnosis of diabetes in a one-year period or one or more hospitalizations with a diagnosis of diabetes. 

 

Changing the case definition resulted in a decrease in Kappa, sensitivity and NPV, and an increase in specificity and PPV

Estimates of the validation indices are included for each case definition 

 

Generally speaking, agreement between the two data sources (as measured by the kappa statistic) was good, but varies with the case definition. With the exception of case definition g (1+H or 2+P, 3 years of data), using 3 years of data did not increase the kappa statistic, when compared to 2 years of data.  Sensitivity was high for all case definitions: all but one value was over 75%.  Sensitivity increased marginally when additional years of data were used; unlike Kappa, this was true for 1,2 and 3 years of data.  Specificity was very high (>97%) for all case definitions, and showed very little variation.  PPV showed a greater range than either sensitivity or specificity. NPV, similar to sensitivity, was very high for all case definitions.