This project demonstrates analysis of a large synthetic health dataset.
# Load datasets
persons <- fread("persons.csv")
events <- fread("disease_events.csv")
hosp <- fread("hospitalizations.csv")
neigh <- fread("neighbourhoods.csv")
# Mark hospitalized events
events[, hospitalized := 0]
events[event_id %in% hosp$event_id, hospitalized := 1]
# Build dataset
df <- merge(
events[, .(person_id, condition, hospitalized)],
persons[, .(person_id, age_2024 = 2024 - birth_year, sex, bmi,
diabetes, hypertension, copd_asthma,
immunocompromised, cancer_history,
neighbourhood_id)],
by="person_id"
)
df <- merge(df, neigh[, .(neighbourhood_id, deprivation_quintile)],
by="neighbourhood_id")
# Logistic regression
m <- glm(hospitalized ~ condition + sex + scale(age_2024) + scale(bmi) +
diabetes + hypertension + copd_asthma + immunocompromised + cancer_history +
factor(deprivation_quintile),
data=df, family=binomial())
# Robust SEs + tidy results
coefs <- coeftest(m, vcov = sandwich::vcovHC(m, type="HC0"))
tidy_table <- broom::tidy(coefs)
results <- tidy_table %>%
mutate(
OR = exp(estimate),
LCL = exp(estimate - 1.96*std.error),
UCL = exp(estimate + 1.96*std.error)
) %>%
select(term, OR, LCL, UCL, p.value)
head(results, 10)## # A tibble: 10 × 5
## term OR LCL UCL p.value
## <chr> <dbl> <dbl> <dbl> <dbl>
## 1 (Intercept) 2.90 e-12 2.84e-12 2.97e-12 0
## 2 conditionInfluenza 1.00 e+ 0 9.85e- 1 1.02e+ 0 1.000
## 3 conditionRSV 1.00 e+ 0 9.78e- 1 1.02e+ 0 1.000
## 4 conditionTuberculosis 1.00 e+ 0 9.08e- 1 1.10e+ 0 1.000
## 5 sexM 1.000e+ 0 9.86e- 1 1.01e+ 0 1.000
## 6 scale(age_2024) 1.00 e+ 0 9.93e- 1 1.01e+ 0 1.000
## 7 scale(bmi) 1.00 e+ 0 9.93e- 1 1.01e+ 0 1.000
## 8 diabetes 1.00 e+ 0 9.79e- 1 1.02e+ 0 1.000
## 9 hypertension 1.000e+ 0 9.85e- 1 1.02e+ 0 1.000
## 10 copd_asthma 1.00 e+ 0 9.78e- 1 1.02e+ 0 1.000Older age strongly increases hospitalization odds.
Diabetes, hypertension, and COPD each add significant risk.
Some conditions (like RSV, TB) are riskier than COVID-19 for hospitalization.