Impact of Hearing impairment and SDOH on cognitive function
semere
2025-01-11
1 Introduction
Research Question: • Is there a difference in the association between cognitive function and hearing aid use between Black and non-Hispanic White older adults, after adjusting for age, education, and Area Deprivation Index (ADI)? • Objective: Find the difference in the association between cognitive function and hearing aid use between Black and non-Hispanic White (nHW) participants using the combined baseline data for ARCHES and DRIVES. ## Statistical Analysis plan: 1. Descriptive Statistics: Perform descriptive statistics for demographic and baseline variables across the two groups (Black and nHW), including education, sex, age, and ADI. 2. Group Comparisons: Use t-tests (for continuous variables) and Chi-square tests (for categorical variables) to compare baseline characteristics between Black and nHW participants. 3. General Linear Model (GLM): Use a general linear model analysis to examine the difference in cognitive function (using PACC score) between Black and nHW participants while controlling for other covariates. Formula: (PACC=β0+β1(Race)+β2(HearingAid)+β3(Race×HearingAid)+β4(Education)+β5(Sex)+β6(Age)+β7(ADI)+ε)
1.1 Visual Summaries of Key Variables
1.1.1 Histograms by Race
1.1.2 Scatter Plots with Regression Lines
### Interpretation of Visuals PACC_BL decreases slightly with age,
increases with education, and strongly correlates positively with WRAT
raw word scores. It shows no clear relationship with ADI_NATRANK and a
weak negative correlation with HHI total scores. All have linear
relationship.
2 Descriptive Statistics (Table 1)
| Total (N=703) |
Black or African American (N=232) |
White (N=471) |
|
|---|---|---|---|
| age | |||
| Mean (SD) | 72.4 (5.05) | 72.1 (5.10) | 72.6 (5.02) |
| Median [Min, Max] | 71.8 [65.1, 94.7] | 71.5 [65.1, 92.9] | 71.9 [65.2, 94.7] |
| gender | |||
| Female | 412 (58.6%) | 177 (76.3%) | 235 (49.9%) |
| Male | 291 (41.4%) | 55 (23.7%) | 236 (50.1%) |
| educ | |||
| Mean (SD) | 16.1 (2.60) | 15.5 (2.81) | 16.4 (2.44) |
| Median [Min, Max] | 16.0 [8.00, 29.0] | 16.0 [8.00, 29.0] | 16.0 [10.0, 24.0] |
| mocatots | |||
| Mean (SD) | 24.9 (3.32) | 24.6 (3.12) | 25.2 (3.43) |
| Median [Min, Max] | 26.0 [11.0, 30.0] | 25.0 [11.0, 30.0] | 26.0 [12.0, 30.0] |
| Missing | 196 (27.9%) | 21 (9.1%) | 175 (37.2%) |
| PACC_BL | |||
| Mean (SD) | -0.0155 (0.729) | -0.325 (0.720) | 0.156 (0.677) |
| Median [Min, Max] | 0.0866 [-2.66, 1.58] | -0.273 [-2.52, 1.27] | 0.245 [-2.66, 1.58] |
| Missing | 101 (14.4%) | 17 (7.3%) | 84 (17.8%) |
| ADI_NATRANK | |||
| Mean (SD) | 54.7 (26.9) | 74.6 (22.7) | 43.5 (22.3) |
| Median [Min, Max] | 52.0 [3.00, 100] | 80.0 [12.0, 100] | 42.0 [3.00, 96.0] |
| Missing | 121 (17.2%) | 23 (9.9%) | 98 (20.8%) |
| wrat_rawword | |||
| Mean (SD) | 62.2 (6.21) | 58.9 (6.78) | 63.9 (5.16) |
| Median [Min, Max] | 63.0 [31.0, 70.0] | 60.0 [31.0, 70.0] | 65.0 [40.0, 70.0] |
| Missing | 2 (0.3%) | 0 (0%) | 2 (0.4%) |
| phq_totscore | |||
| Mean (SD) | 4.32 (4.92) | 3.60 (4.46) | 4.88 (5.18) |
| Median [Min, Max] | 3.00 [0, 22.0] | 2.00 [0, 20.0] | 3.00 [0, 22.0] |
| Missing | 274 (39.0%) | 45 (19.4%) | 229 (48.6%) |
| PACC_class_sick_65 | |||
| high PACC | 515 (73.3%) | 162 (69.8%) | 353 (74.9%) |
| low PACC | 87 (12.4%) | 53 (22.8%) | 34 (7.2%) |
| Missing | 101 (14.4%) | 17 (7.3%) | 84 (17.8%) |
| noise_censusblock2020_mean | |||
| Mean (SD) | 52.2 (2.64) | 51.5 (2.85) | 52.5 (2.46) |
| Median [Min, Max] | 52.2 [45.8, 60.1] | 51.5 [45.8, 59.6] | 52.6 [46.0, 60.1] |
| Missing | 5 (0.7%) | 3 (1.3%) | 2 (0.4%) |
| Hearing Aid Use | |||
| No | 599 (85.2%) | 212 (91.4%) | 387 (82.2%) |
| Yes | 104 (14.8%) | 20 (8.6%) | 84 (17.8%) |
| MOCA Category | |||
| high moca | 303 (43.1%) | 142 (61.2%) | 161 (34.2%) |
| low moca | 204 (29.0%) | 69 (29.7%) | 135 (28.7%) |
| Missing | 196 (27.9%) | 21 (9.1%) | 175 (37.2%) |
2.1 Interpretation of Table 1
Blacks are slightly younger, have lower education levels, and higher ADI National Rank compared to White participants. MOCA Total Scores, WRAT Raw Scores, and PHQ Total Scores are generally higher in White participants. Blacks have a higher mean ADI National Rank (73.8) compared to White participants (42.3)and Hearing aid use is reported by 8.6% of Blacks and 17.8% of Whites.
3 GLM with PACC score and MOCA (continous) as the outcome
3.0.1 GLM with interaction term and main effect
Interaction term and hearing aid use were not significant, only race was a significant predictor for both MOCA and PACC| Predictor | PACC Estimate | PACC Std. Error | PACC CI | PACC P-Value | MOCA Estimate | MOCA Std. Error | MOCA CI | MOCA P-Value |
|---|---|---|---|---|---|---|---|---|
| (Intercept) | -0.3352626 | 0.0493420 | [-0.43, -0.24] | <0.001 | 24.5876289 | 0.2369999 | [24.12, 25.05] | <0.001 |
| race_releveledWhite | 0.5100990 | 0.0628683 | [0.39, 0.63] | <0.001 | 0.7852525 | 0.3199090 | [0.16, 1.41] | 0.0144 |
| hearingaids1_releveledYes | 0.1237046 | 0.1705295 | [-0.21, 0.46] | 0.468 | 0.0594300 | 0.8349586 | [-1.58, 1.7] | 0.9433 |
| race_releveledWhite:hearingaids1_releveledYes | -0.2239681 | 0.1932701 | [-0.6, 0.15] | 0.247 | -0.9323113 | 0.9617389 | [-2.82, 0.95] | 0.3328 |
3.0.2 GLM with all relevant varaibles
| Predictor | PACC Estimate | PACC Std. Error | PACC CI | PACC P-Value | MOCA Estimate | MOCA Std. Error | MOCA CI | MOCA P-Value |
|---|---|---|---|---|---|---|---|---|
| (Intercept) | 2.0128557 | 0.7907950 | [0.46, 3.56] | 0.01122 | 33.5393529 | 4.1510038 | [25.4, 41.68] | <0.001 |
| race_releveledWhite | 0.5029635 | 0.0786523 | [0.35, 0.66] | < 0.001 | 0.4793237 | 0.4100539 | [-0.32, 1.28] | 0.2431 |
| hearingaids1_releveledYes | 0.0528035 | 0.0895762 | [-0.12, 0.23] | 0.55581 | -0.1088015 | 0.4810450 | [-1.05, 0.83] | 0.8212 |
| ADI_NATRANK | -0.0008071 | 0.0014497 | [0, 0] | 0.57797 | -0.0047617 | 0.0077220 | [-0.02, 0.01] | 0.5378 |
| noise_censusblock2020_mean | -0.0024592 | 0.0117450 | [-0.03, 0.02] | 0.83424 | 0.0667506 | 0.0619405 | [-0.05, 0.19] | 0.2818 |
| educ | 0.0333378 | 0.0121785 | [0.01, 0.06] | 0.00641 | -0.0259526 | 0.0647857 | [-0.15, 0.1] | 0.6889 |
| age | -0.0364597 | 0.0059504 | [-0.05, -0.02] | < 0.001 | -0.1573990 | 0.0316056 | [-0.22, -0.1] | <0.001 |
| genderMale | -0.2024528 | 0.0661706 | [-0.33, -0.07] | 0.00234 | -0.7776095 | 0.3603758 | [-1.48, -0.07] | 0.0315 |
3.0.3 minimum model with out interaction term
Hearing aid use was not significant (P= 0.1211).| Predictor | PACC Estimate | PACC Std. Error | PACC CI | PACC P-Value | MOCA Estimate | MOCA Std. Error | MOCA CI | MOCA P-Value |
|---|---|---|---|---|---|---|---|---|
| (Intercept) | -0.3206648 | 0.0477205 | [-0.41, -0.23] | <0.001 | 24.6442454 | 0.2296773 | [24.19, 25.09] | <0.001 |
| race_releveledWhite | 0.4864005 | 0.0594662 | [0.37, 0.6] | <0.001 | 0.6820953 | 0.3016738 | [0.09, 1.27] | 0.0242 |
| hearingaids1_releveledYes | -0.0506589 | 0.0802771 | [-0.21, 0.11] | 0.528 | -0.6432807 | 0.4143288 | [-1.46, 0.17] | 0.1211 |
4 Summary stat for pacc bin and noise data based on census block mean
PACC bin was created by taking those less than -1 sd in as low pacc and those above -1 sd as high pacc| PACC Class | Count |
|---|---|
| high PACC | 515 |
| low PACC | 87 |
| NA | 101 |
| Mean | Median | SD | Min | Max | Observations |
|---|---|---|---|---|---|
| 52.19714 | 52.16267 | 2.637775 | 45.82583 | 60.08316 | 698 |
5 Mediation analysis
Race as independent variable, hearing aid use as the mediator and PACC scrore as the outcome variable. It was done via The Imai–Keele–Tingley (IKT) causal mediation framework, which is a general approach to causal mediation analysis that can accommodate a variety of statistical models.
5.1 Table Explanation
ACME (Average Causal Mediation Effect): 0.005 (95% CI includes 0), p = 0.48, indicating no significant mediation through hearing aid use. ADE (Average Direct Effect): -0.486 (95% CI excludes 0), p < 2e-16, showing a strong negative direct effect of race on PACC. Total Effect: -0.482, very similar to the ADE, implying virtually all of the effect is direct(race to pacc). Prop. Mediated: ~ -1%, p = 0.48, also non-significant, confirming negligible mediation. ## Plot Explanation The forest-style plot displays the three main effect estimates (ACME, ADE, Total Effect) with their 95% confidence intervals. ACME is centered near zero with a wide interval overlapping zero → not significant. ADE and Total Effect both lie well below zero with narrow confidence intervals → strongly significant. Visually, it’s clear the mediated portion (ACME) is negligible compared to the direct effect and total effect. Overall, hearing aid use does not appear to mediate the relationship between race and PACC scores, as almost the entire negative association of race with PACC is direct (i.e., not explained by the mediator).
#selecting variables for mediation analysis
mediation_analysis_vars <- c("race_releveled", "hearingaids1_releveled", "PACC_BL")
#Filter data for complete cases
clean_mediation_data <- combined_Drives_Arches_above_65[
complete.cases(combined_Drives_Arches_above_65[mediation_analysis_vars]),
]
#1. Make a fresh copy of cleaned dataset
data_for_mediation <- clean_mediation_data
#2. Recode 'race_releveled' into numeric 0/1
data_for_mediation$race_binary <- ifelse(
data_for_mediation$race_releveled == "White", 0, 1
)
#3. Recode 'hearingaids1_releveled' into numeric 0/1
data_for_mediation$ha_binary <- ifelse(
data_for_mediation$hearingaids1_releveled == "No", 0, 1
)
#4. Fit mediator model with numeric mediator
mediator_model_simple <- glm(
ha_binary ~ race_binary,
family = binomial(link = "logit"),
data = data_for_mediation
)
# 5. Fit outcome model (linear) with numeric treatment & mediator
outcome_model_simple <- lm(
PACC_BL ~ race_binary + ha_binary,
data = data_for_mediation
)
#6. Mediate (now treat & mediator are numeric)
mediation_result_simple <- mediate(
model.m = mediator_model_simple,
model.y = outcome_model_simple,
treat = "race_binary",
mediator = "ha_binary",
treat.value = 1,
control.value = 0,
boot = TRUE, # or FALSE
sims = 1000
)| Effect | Estimate | 95% CI | p_value |
|---|---|---|---|
| ACME | 0.005 | [-0.00812, 0.02] | 0.48 |
| ADE | -0.486 | [-0.60908, -0.37] | <2e-16 |
| Total Effect | -0.482 | [-0.60169, -0.36] | <2e-16 |
| Prop. Mediated | -0.010 | [-0.04913, 0.02] | 0.48 |
6 AIM 2 analysis
6.1 Objective:
Build a logistic regression model to show the association among sound pollution level in dB, ADI, race/ethnicity, and hearing aid use to predict PACC score bin or MOCA while controlling for education, age, gender on the combined dataset of ARCHES and Mother. ## Method: 1. Logistic Regression Model: Build a logistic regression model to predict either the PACC score bin and the MOCA. Predictors: Include sound pollution level in DB, ADI, race/ethnicity, hearing aid use as predictors. Control Variables: Include education, age, gender as covariates in the model to control for their potential influence. Model Evaluation: Use metrics such as AUC (Area Under the Curve), Hosmer-Lemeshow test, and classification accuracy to evaluate the performance of the logistic regression model. Exclusion criteria Participants with missing data for PACC score or MOCA classification.
# MOCA_Cat Model
moca_model_data <- moca_model_data %>%
mutate(MOCA_Cat = factor(MOCA_Cat, levels = c("low moca", "high moca")))
moca_model_logistic <- glm(
MOCA_Cat ~ noise_censusblock2020_mean + ADI_NATRANK + race + hearingaids1 +
educ + age + gender,
family = binomial(link = "logit"),
data = moca_model_data
)
pacc_model_data$PACC_class_sick_65 <- factor(
pacc_model_data$PACC_class_sick_65,
levels = c("low PACC", "high PACC")
)
# PACC_class_sick_65 Model
pacc_model_logistic <- glm(
PACC_class_sick_65 ~ noise_censusblock2020_mean + ADI_NATRANK + race + hearingaids1 +
educ + age + gender,
family = binomial(link = "logit"),
data = pacc_model_data
)
# Step 3: Model Evaluation Function
evaluate_model <- function(model, data, outcome) {
# Generate predictions
predictions <- predict(model, type = "response")
# ROC and AUC
roc_obj <- roc(data[[outcome]], predictions, levels = levels(data[[outcome]]))
auc_value <- auc(roc_obj)
# Hosmer-Lemeshow Test
hoslem_test <- hoslem.test(as.numeric(data[[outcome]]) - 1, predictions, g = 10)
# Confusion Matrix
predicted_classes <- ifelse(predictions > 0.5, levels(data[[outcome]])[2], levels(data[[outcome]])[1])
confusion_matrix <- table(Predicted = predicted_classes, Actual = data[[outcome]])
list(
AUC = auc_value,
HosmerLemeshow = hoslem_test,
ConfusionMatrix = confusion_matrix
)
}
# Step 4: Evaluate Models
moca_results_logistic <- evaluate_model(moca_model_logistic, moca_model_data, "MOCA_Cat")
pacc_results_logistic <- evaluate_model(pacc_model_logistic, pacc_model_data, "PACC_class_sick_65")
# Step 5: Summarize Results
results_logistic_regression <- list(
MOCA_Model_Summary = summary(moca_model_logistic),
MOCA_Evaluation = moca_results_logistic,
PACC_Model_Summary = summary(pacc_model_logistic),
PACC_Evaluation = pacc_results_logistic
)| Predictor | MOCA OR | MOCA 95% CI | MOCA p-value | PACC OR | PACC 95% CI | PACC p-value |
|---|---|---|---|---|---|---|
| (Intercept) | 6109.664 | [28.397, 1544197.111] | 0.00168 | 1381.737 | [2.268, 947253.735] | 0.02810 |
| noise_censusblock2020_mean | 1.016 | [0.938, 1.102] | 0.69431 | 1.023 | [0.929, 1.127] | 0.64744 |
| ADI_NATRANK | 0.991 | [0.981, 1.001] | 0.06878 | 0.993 | [0.98, 1.005] | 0.25396 |
| raceWhite | 0.445 | [0.258, 0.758] | 0.00321 | 2.528 | [1.327, 4.88] | 0.00511 |
| hearingaids1Yes | 0.792 | [0.435, 1.451] | 0.44721 | 1.391 | [0.656, 3.221] | 0.41235 |
| educ | 0.974 | [0.896, 1.059] | 0.53056 | 1.146 | [1.036, 1.273] | 0.00937 |
| age | 0.902 | [0.863, 0.94] | < 0.001 | 0.887 | [0.846, 0.928] | < 0.001 |
| genderMale | 0.732 | [0.463, 1.158] | 0.18081 | 0.675 | [0.388, 1.178] | 0.16402 |
| Model | AUC | Hosmer.Lemeshow.p.value |
|---|---|---|
| MOCA Model | 0.686 | 0.648 |
| PACC Model | 0.732 | 0.0115 |
6.1.1 MoCA Model
Key findings: Race (White) and older age significantly predicted higher MoCA status, while female gender was associated with lower MoCA status. Education was marginally negative (p=0.07), contrary to expectations. Model performance: AUC = 0.67 (moderate discrimination), Hosmer–Lemeshow p=0.54 (acceptable calibration). Census-block noise and hearing-aid use were not significant.
6.1.2 PACC Model
Key findings: Higher education and being White reduced the odds of low PACC, while older age increased that risk. Model performance: AUC = 0.74 (good discrimination), Hosmer–Lemeshow p=0.016 (some calibration concern). Census-block noise and hearing-aid use were not significant.