Estimate Per-Class LOS Relative Risk via Quasi-Poisson Regression

For each pathogen k and antibiotic class c, fits: log(E[LOS_i]) = b0 + b_c * class_ci + b_HAI * HAI_i + sum_f gf * I(centre=f) and returns RR_kc_LOS = exp(b_c) with 95

Usage

fit_los_rr_poisson(
  data,
  patient_id_col = "PatientInformation_id",
  isolate_id_col = "isolate_id",
  facility_col = "center_name",
  organism_col = "organism_name",
  syndrome_col = "syndrome",
  infection_type_col = "type_of_infection",
  antibiotic_class_col = "antibiotic_class",
  antibiotic_name_col = "antibiotic_name",
  antibiotic_value_col = "antibiotic_value",
  date_admission_col = "date_of_admission",
  date_discharge_col = "final_outcome_date",
  date_culture_col = "date_of_first_positive_culture",
  final_outcome_col = "final_outcome",
  final_outcome_value = "Discharged",
  syndrome_name = NULL,
  organism_name = NULL,
  hai_threshold_hours = 48,
  centre_model = c("pooled_fe", "per_centre"),
  pool_method = c("weighted_mean", "median"),
  max_los = 365,
  min_n = 10L
)

Arguments

data

Data frame. Merged AMR dataset at isolate x antibiotic level.

patient_id_col

Character. Default "PatientInformation_id".

isolate_id_col

Character. Default "isolate_id".

facility_col

Character. Default "center_name".

organism_col

Character. Default "organism_name".

syndrome_col

Character. Default "syndrome".

infection_type_col

Character. Default "type_of_infection".

antibiotic_class_col

Character. Default "antibiotic_class".

antibiotic_name_col

Character. Default "antibiotic_name".

antibiotic_value_col

Character. Default "antibiotic_value".

date_admission_col

Character. Default "date_of_admission".

date_discharge_col

Character. Default "final_outcome_date".

date_culture_col

Character. Default "date_of_first_positive_culture".

final_outcome_col

Character. Default "final_outcome".

final_outcome_value

Character. Default "Discharged".

syndrome_name

Character or NULL. Filter to syndrome (e.g. "BSI"). NULL = all syndromes (universal RR).

organism_name

Character or NULL. Pathogen(s) to process. NULL = all.

hai_threshold_hours

Numeric. HAI/CAI gap threshold. Default 48.

centre_model

Character. "pooled_fe" (default) or "per_centre".

pool_method

Character. Used when centre_model = "per_centre". "weighted_mean" (default) or "median".

max_los

Numeric. Default 200.

min_n

Integer. Minimum patients per model. Default 10.

Value

Data frame: pathogen, antibiotic_class, RR_LOS, CI_lower, CI_upper, n_patients, centre_model, syndrome_scope. When centre_model = "per_centre", attribute "per_centre_rr" contains per-centre estimates.

Details

LOS is computed with HAI/CAI-specific clock (see compute_patient_los). Resistance is classified at antibiotic class level (class = 1 if any drug in class = R). HAI enters as a binary covariate to absorb residual baseline severity differences beyond the LOS clock adjustment. No PreDays covariate is used – the LOS clock change from admission to culture date for HAI patients already removes the pre-infection period from the outcome.

Stated limitation: when syndrome_name is supplied, RR_kc_LOS is syndrome-specific. Downstream PAF applies this RR to all profiles of pathogen k, assuming syndrome-invariant LOS prolongation across infection sources. Set syndrome_name = NULL for a universal RR.