An online experiment to assess bias in professional medical coding

Table 3 Ordinary least squares regression models of level of service score (range: 1–5) assigned to six medical visit scenarios by professional medical coders, by patient demographic characteristics^*

	Chart 1		Chart 2		Chart 3		Chart 4		Chart 5		Chart 6
	Coef	SE	Coef	SE	Coef	SE	Coef	SE	Coef	SE	Coef	SE
Pooled Experimental Arm 1 (Racial Bias)
White patient or no race identified (ref)
African-American patient	0.009	(0.10)	0.067	(0.10)	0.114	(0.09)	−0.008	(0.10)	0.088	(0.11)	0.179	(0.13)
Pooled Experimental Arm 2 (Age Bias)
Middle-aged patient (ref)
Older adult patient	−0.177	(0.11)	0.199	(0.11)*	0.238	(0.09)**	−0.025	(0.10)	0.076	(0.09)	−0.094	(0.14)
Pooled Experimental Arm 3 (Ability Bias)
Patient with no disabilities (ref)
Patient with hearing/visual/physical disability	0.048	(0.11)	0.044	(0.10)	0.117	(0.08)	0.100	(0.10)	0.177	(0.10)*	−0.074	(0.14)
Pooled Experimental Arm 4 (Gender Bias)^†
Male patient (ref)
Female patient	−0.175	(0.10)*	0.015	(0.10)	−0.176	(0.09)**	0.214	(0.10)**	−0.037	(0.10)	0.175	(0.13)
Pooled Experimental Arm 5 (Social Need)
No social need (ref)
Any social need	0.097	(0.08)	0.045	(0.08)	0.081	(0.07)	0.009	(0.08)	−0.042	(0.07)	0.023	(0.10)

Source: Original data from an online experiment of professional medical coders in the US, August–September, 2017. This table presents tests that pooled responses across study arms when the clinical scenarios were identical. *Models control for respondent gender, age in years, race/ethnicity, and years worked as a professional medical coder. † Chart 6 contrasts a female patient to a patient with no identified gender *p < 0.10, **p < 0.05

ISSN: 1472-6947