Skip to main content

Table 3 Summary output of Bayesian geostatistical models for the north and south of Somalia.

From: Spatial prediction of Plasmodium falciparum prevalence in Somalia

Model/Variables

North

South

Bayesian geostatistical model (no covariates)

  

α (Intercept)

-4.62 (-5.44, -4.33)

-2.9 (-3.37, -2.27)

Ï• (Decay of spatial correlation (degrees latitude and longitude))

8.90 (3.11, 12.75)

4.79 (2.11, 6.97)

σ2 (Variance of spatial process)

4.35 (271, 7.14)

7.14 (5.00, 8.76)

DIC

326

1,454

ME (% Pf PR)

3.83

4.14

MAE (% Pf PR)

4.12

5.06

AU-ROC*

  

   <5% Pf PR

0.72 (0.64, 0.86)

0.87 (0.72, 0.91)

   5–39% Pf PR

0.66 (0.51, 0.80)

0.78 (0.66, 0.85)

   ≥ 40% Pf PR

NA

0.56 (0.37, 0.73)

Bayesian geostatistical model (with covariates)

  

α (Intercept)

-4.62 (-5.23, -4.10)

-2.86 (-3.79, 2.27)

Ï• (Decay of spatial correlation)

10.35 (4.70, 12.88)

5.78 (2.95, 6.99)

σ2 (Variance of spatial process)

3.70 (2.17, 7.14)

5.00 (3.70, 6.70)

DIC

323

1,429

ME (% Pf PR)

2.56

3.65

MAE (% Pf PR)

4.75

5.00

AU-ROC*

  

   <5% Pf PR

0.75 (0.64, 0.91)

0.91 (0.87, 0.99)

   5–39% Pf PR

0.64 (0.43, 0.84)

0.81 (0.70, 0.94)

   ≥ 40% Pf PR

NA

0.51 (0.32, 0.83)

 

Odds ratio, (95% Bayes credible interval)

Month of survey

  

   Feb

Ref

Ref

   Mar

-

4.06 (2.20, 7.63)

   Jun

-

 

   Jul

3.25 (0.91, 11.36)

0.87 (0.48,1.46)

   Sep

0.2 (0.02, 1.74)

 

   Nov

1.31 (0.33, 4.36)

0.48 (0.23, 0.96)

   Dec

-

1.95 (0.91, 3.90)

Annual average minimum temperature

  

   <median of 20.4/22.1 (North/South)°C

Ref

-

   >median of 20.4/22.1(North/South)°C

1.12 (0.84, 1.33)

0.83 (0.67,0.96)

Annual average precipitation

1.70(0.53, 5.44)

1.41 (1.07, 1.94)

Distance to water features (km)

1.22 (0.53, 2.81)

0.79 (0.74, 1.29)

  1. DIC = deviance information criterion (measure of model fit); ME = mean error (measure of overall bias); MAE = mean absolute error (measure of overall precision); AU-ROC = areas under receiver operating characteristic. Values in parentheses are 95% Bayes credible intervals.
  2. *There were only two survey locations in the validation set that had Pf PR ≥ 40% in the north region meaning reliable AU-ROC values could not be computed for this model.