Skip to main content

Table 2 Log-linear regression models to assess risk of symptom(s) at follow-up

From: Burden of long COVID among adults experiencing sheltered homelessness: a longitudinal cohort study in King County, WA between September 2020—April 2022

 

aRR

95% CI

p-value

Primary analysis†

 Model A: No covariatesa

  Case (vs. control)

1.500

(0.949—2.371)

0.083

 Model B: Add covariates time and follow-up season onlyb

  Case (vs. control)

5.365

(2.731—10.539)

 < 0.001*

 Model C: Sensitivity analysis—expanded modelc

  Case (vs. control)

5.735

(1.086—30.285)

0.040*

 Model D: Sensitivity analysis—full modeld

  Case (vs. control)

8.902

(2.015—39.324)

0.004*

Secondary analysis‡

 Model A: No covariatesa

  Case (vs. Control, ORV-positive)

1.799

(0.663—4.902)

0.249

  Case (vs. Control, ORV-negative)

1.285

(0.791—2.092)

0.311

 Model B: Add covariates time and follow-up season onlyb

  Case (vs. Control, ORV-positive)

6.250

(1.805—21.739)

0.004*

  Case (vs. Control, ORV-negative)

4.237

(1.499—11.905)

0.006*

 Model C: Sensitivity analysis—expanded modelc

  Case (vs. Control, ORV-positive)

7.143

(1.206—41.667)

0.030*

  Case (vs. Control, ORV-negative)

4.237

(0.597—30.303)

0.149

 Model D: Sensitivity analysis—full modeld

  Case (vs. Control, ORV-positive)

11.494

(2.045—62.500)

0.006*

  Case (vs. Control, ORV-negative)

6.494

(1.292—32.258)

0.023*

  1. † Primary analysis uses binary exposure (COVID-19-positive case vs. COVID-19-negative control)
  2. ‡ Secondary analysis uses categorical exposure (COVID-19-positive case vs. COVID-19-negative control, ORV-positive vs. COVID-19 control, ORV-negative)
  3. a Model A represents the model with no covariates, adjusted only for age and sex due to frequency matching in study design
  4. b Model B builds upon Model A by also including adjustment for time since enrollment via 4 restricted cubic splines (centered at day 90) and follow-up season (categorical: fall, spring, summer, winter). As we found no evidence of interaction between case status and time since enrollment in either model, the interaction term was not included in the presented results
  5. c Model C builds upon Model B by also including adjustment for key covariates identified a priori, including race (categorical: White, Black/African American, Other), any comorbidities (binary: yes, no), income (binary: ≥ $25,000, < $25,000) and smoking status (binary: smoker, non-smoker), and additional covariates added during bivariate and forward selection screening procedures, including duration of homelessness (categorical: < 12 months, 12 + months, prefer not to say/missing) and employment (binary: yes, no)
  6. d Model D builds upon Model C by also including adjustment for all remaining potential covariates, including insurance (binary: yes, no), education (binary: high school or less, some college or more), and hispanic ethnicity (binary: yes, no)
  7. *p-value < 0.05
Back to article page

BMC Public Health

ISSN: 1471-2458

Contact us