Health Literacy and Hospital Length of Stay: An Inpatient Cohort Study
BACKGROUND: Associations between low health literacy (HL) and adverse health outcomes have been well documented in the outpatient setting; however, few studies have examined associations between low HL and in-hospital outcomes.
OBJECTIVE: To compare hospital length of stay (LOS) among patients with low HL and those with adequate HL.
DESIGN: Hospital-based cohort study.
SETTING: Academic urban tertiary-care hospital.
PATIENTS: Hospitalized general medicine patients.
MEASUREMENTS: We measured HL using the Brief Health Literacy Screen. Severity of illness and LOS were obtained from administrative data. Multivariable linear regression controlling for illness severity and sociodemographic variables was employed to measure the association between HL and LOS.
RESULTS: Among 5540 participants, 20% (1104/5540) had low HL. Participants with low HL had a longer average LOS (6.0 vs 5.4 days, P < 0.001). Low HL was associated with an 11.1% longer LOS (95% confidence interval [CI], 6.1%-16.1%; P < 0.001) in multivariate analysis. This effect was significantly modified by gender (P = 0.02). Low HL was associated with a 17.8% longer LOS among men (95% CI, 10.0%-25.7%; P < 0.001), but only a 7.7% longer LOS among women (95% CI, 1.9%-13.5%; P = 0.009).
CONCLUSIONS: In this single-center cohort study, low HL was associated with a longer hospital LOS. The findings suggest that the adverse effects of low HL may extend into the inpatient setting, indicating that targeted interventions may be needed for patients with low HL. Further work is needed to explore these negative consequences and potential mitigating factors.
© 2017 Society of Hospital Medicine
Statistical Analysis
χ2 and 2-tailed t tests were used to compare categorical and continuous variables, respectively. Multivariate linear regressions were employed to measure associations between the independent variables (HL, illness severity, race, gender, education, and insurance status) and the dependent variable, LOS. Independent variables were chosen for clinical significance and retained in the model regardless of statistical significance. The adjusted R2 values of models with and without the HL variable included were reported to provide information on the contribution of HL to the overall model.
Because LOS was observed to be right skewed and residuals of the untransformed regression were observed to be non-normally distributed, the decision was made to natural log transform LOS, which is consistent with previous hospital LOS studies.16 Regression coefficients and confidence intervals were then transformed into percentage estimates using the following equation: 100(eβ–1). Adjusted R2 was reported for the transformed regression.
The APR-DRG relative weight was treated as a continuous variable. Sociodemographic variables were dichotomized as follows: female vs male; high school graduates vs not; African American vs not; Medicaid/no payer vs Medicare/private payer. Age was not included in the multivariate model because it has been incorporated into the weighted APR-DRG illness severity scores.
Each of the sociodemographic variables and the APR-DRG score were examined for effect modification via the same multivariate linear equation described above, with the addition of an interaction term. A separate regression was performed with an interaction term between age (dichotomized at ≥65) and HL to investigate whether age modified the association between HL and LOS. Finally, we explored whether effects were isolated to long vs short LOS by dividing the sample based on the mean LOS (≥6 days) and performing separate multivariate comparisons.
Sensitivity analyses were performed to exclude those with LOS greater than the 90th percentile and those with APR-DRG score greater than the 90th percentile; age was added to the model as a continuous variable to evaluate whether the illness severity score fully adjusted for the effects of age on LOS. Furthermore, we compared the participants with missing data to those with complete data across both dependent and independent variables. Alpha was set at 0.05; analyses were performed using Stata Version 14 (Stata, College Station, TX).
RESULTS
A total of 5983 participants met inclusion criteria and completed the HL assessment; of these participants, 75 (1%) died during hospitalization, 9 (0.2%) had missing discharge status, and 79 (1%) had LOS >30 days. Two hundred eighty (5%) were missing data on sociodemographic variables or APR-DRG score. Of the remaining (n = 5540), the mean age was 57 years (standard deviation [SD] = 19 years), over half of participants were female (57%), and the majority were African American (73%) and had graduated from high school (81%). The sample was divided into those with private insurance (25%), those with Medicare (46%), and those with Medicaid (26%); 2% had no payer. The mean APR-DRG score was 1.3 (SD = 1.2), and the scores ranged from 0.3 to 15.8.
On the BHLS screen for HL, 20% (1104/5540) had inadequate HL. Participants with low HL had higher weighted illness severity scores (average 1.4 vs 1.3; P = 0.003). Participants with low HL were also more likely to be 65 or older (55% vs 33%; P < 0.001), non-high school graduates (35% vs 15%; P < 0.001), and African American (78% vs 72%; P < 0.001), and to have Medicare or private insurance (75% vs 71%; P = 0.02). There was no significant difference with respect to gender (54% male vs 57% female; P = 0.1)
Finally, we compared the group with missing data (n = 280) to the group with complete data (n = 5540). The participants with missing data were more likely to have low HL (31% [86/280] vs 20%; P < 0.001) and to have Medicare or private insurance (82% [177/217] vs 72%; P = 0.002); however, they were not more likely to be 65 or older (40% [112/280] vs 37%; P = 0.3), high school graduates (88% [113/129] vs 81%; P = 0.06), African American (69% [177/256] vs 73%; P = 0.1), or female (57% [158/279] vs 57%; P = 1), nor were they more likely to have longer LOS (5.7 [n = 280] vs 5.5 days; P = 0.6) or higher illness severity scores (1.3 [n = 231] vs 1.3; P = 0.7).
