Skip to main content
. 2021 Jun 23;129(6):067004. doi: 10.1289/EHP7813

Table 5.

Multivariable modeling of land use and bedrock risk factors as related to detection probabilities and concentrations of genetic markers of host-specific and fecal-associated microbes in private wells.

Fecal marker source and outcome measurement (n)a Risk factor Univariable model
p-value
Multivariable model
Risk factor medianb Risk factor rangeb Coefficient or trendc,d p-valuee
Human marker detection (137) Drainfield septic systems, count within 229m 0.038 2 0–10 0.09 0.11
Groundwater depth, 14-d antecedent, median 0.0003 1.2 0.3–1.6 0.9 0.011
Rainfall, 2-d antecedent, cumulative 0.0093 14 0–37 Positive 0.69
Bedrock depth 0.051 6.1 0–46.6 Negative 0.13
Human marker concentration (33) Bedrock depth 0.011 4.3 0.3–36.6 Negative 0.011
Bovine marker detection (138) Groundwater recharge, 7-d antecedent, cumulative 0.0041 50 0–60 Positive 0.0092
Bovine marker concentration (41) Agricultural field area within 229m 0.029 11.6 3.7–16.4 Positive 0.024
Bedrock depth 0.0019 5.2 0–29 0.1 0.0006
Any fecal markerf detection (137) Drainfield septic systems, count within 229m 0.0036 2 0–10 Positive 0.036
Rainfall, 2-d antecedent, cumulative 0.12 14 0–37 Positive 0.19
Manure storage distanceg 0.94 687 71–3,728 0.0004 0.036
Bedrock depth 0.027 6.1 0–46.6 0.06 0.0058
Manure storage distance times bedrock depth interaction 0.045 NA NA Negative 0.024
Any fecal marker concentration (77) Agricultural field area within 229m 0.035 12.7 1.1–16.4 Positive 0.097
Manure storage distance 0.083 762 113–3,728 0.0001 0.76
Bedrock depth 0.0003 4.6 0–36.6 0.08 0.002

Note: NA, Not applicable. Univariable model p-values used for selecting risk factors are included for reference; complete univariable statistics are provided in Table S10. Risk factor eligibility for inclusion in multivariable models is described in statistical methods.

a

n=number of samples in multivariable model.

b

Units for distance and depth are meters; rainfall and recharge are millimeters; area is hectares.

c

In lieu of reporting multiple coefficients for spline-represented variables we report the overall trend (positive or negative).

d

Interpretation of coefficient linear terms: change in ln(detection probability) or change in ln(concentration) for a unit change in the risk factor.

e

The composite p-value is reported for spline-represented variables.

f

“Any fecal marker” includes all microorganisms regardless of host specificity.

g

Included in multivariable model because of its significant interaction with bedrock depth.