En Randy Ellis ja abans de la pandèmia estava apuntant cap un nou sistema que permetés la mesura de la morbiditat amb dades diagnòstiques sense una classificació categòrica dels pacients. És a dir es tracta d'una evolució dels models que ell ha proposat des de fa anys junt amb Arlene Ash, els DCGs, i alhora una evolució del Clinical Classification Software.
L'objectiu:
Our objective was to create a clinically detailed, transparent, well-documented, nonproprietary classification system suitable for predicting diverse outcomes using ICD-10-CM diagnostic information and share a core set of predictive models that can be used on other data sets and populations.
Els tres tipus:
We created 3 types of DXIs. The primary or main effect DXIs, called DXI_1, focus on clinical dimensions in each diagnosis. Diagnoses were assigned up to 4 DXI_1s. In some cases, we created both broader and narrower DXI_1s that overlapped because we did not know a priori the level of detail preferred for prediction. We illustrate this approach below in our discussion of sepsis and hypertension in pregnancy DXI_1s.
The second group, DXI_2 modifiers, cut across DXI_1s. Some identify disease severity, such as “with complications,” “hemorrhage,” “secondary,” “bilateral,” and “with coma.” Others may be useful for disease monitoring, including flags for future research and epidemiological surveillance, such as sexually transmitted and vaccine-preventable infectious diseases. Certain diagnoses for external causes and factors influencing health status (whose codes begin with V-Z) were not assigned a DXI_1 and were instead only assigned DXI_2 modifiers.
Finally, DXI_3 scaled variables capture test results, disease severity, or clinically relevant distinctions not easily captured in binary DXI_1 categories. These include body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), neonatal birth weight, neonatal gestational age, pregnancy trimester, low vision/blindness stages, coma scale measures, stroke scores, and duration of unconsciousness. As an example, the DXI_3 variable for BMI, calculated as weight in kilograms divided by height in meters squared, takes on values between 18.5 and 70, corresponding to ordered groups of BMI ranges. When comparing the DXI classification system to existing models, we included only main effects (DXI_1s) as predictors. This comparison cleanly demonstrates the value of the DXIs richer classification of diagnoses. Quantifying the additional value of using DXI_2 and DXI_3 items is left for future research.
Exemples:
La capacitat predictiva en despesa, aquí:
Si això és així, caldrà provar-ho ben aviat, amb 2929 variables explicatives assoleix una variació explicada del 51%.
El software per provar-lo en SAS aquí.
