Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 107

Table 22. Standard error for coefficients for normal compared to gamma (negative binomial)

Coefficient Normal Gamma

Age 0.0005 0.0001

Female 0.0232 0.0038

Hemodialysis 0.0815 0.0133

Continuous Ventilation > 96 Hours 0.1570 0.0245

Insertion of endotracheal tube 0.1533 0.0238

Injection of antibiotic 0.0725 0.0117

Respiratory medication administered by nebulizer 0.0795 0.0128

Diagnostic ultrasound of heart 0.0908 0.0148

Esophagogastroduodenoscopy with closed biopsy 0.1119 0.0180

Continuous positive airway pressure 0.1158 0.0188

Closed endoscopic biopsy of lung 0.1377 0.0216

Computerized axial tomography of head 0.1028 0.0166

Transfusion of packed cells 0.0558 0.0090

Venous catheterization 0.0638 0.0100

Continuous Ventilation < 96 Hours 0.1619 0.0255

Closed biopsy of bronchus 0.0802 0.0125

Thoracentesis 0.0868 0.0140

Quartile 1 0.0335 0.0055

Quartile 2 0.0340 0.0055

Quartile 3 0.0346 0.0056

Caucasian 0.0778 0.0126

African American 0.0847 0.0137

Hispanic 0.0852 0.0138

Asian 0.1136 0.0186

Native American 0.1883 0.0304

future trends

As more complex models come into more common usage in healthcare, they will be adapted to the
development of patient severity indices. In particular, the generalized linear model has features that are
particularly relevant to healthcare data, namely non-normal distributions. We cannot continue to force
the data to fit the model; we must find models that fit the data because the data assumptions are valid.
In the past, models have been used regardless of their applicability in terms of assumptions. The results
of such studies are highly questionable. There are, in fact, many models that have been developed fairly
recently that can be useful in health outcomes research. In chapter 4, we will show some additional
models that are relevant to the analysis of healthcare data.