Skip to main content

Table 5 Performance of predictive generalised linear models in Somalia for real-time estimation, by acute malnutrition outcome

From: Can we predict the burden of acute malnutrition in crisis-affected countries? Findings from Somalia and South Sudan

Statistic

Categorical outcomes

Continuous outcomes

SAM (WFH + oedema)

GAM (WFH + oedema)

SAM (MUAC + oedema)

GAM (MUAC + oedema)

WFH

MUAC

Predictors: coefficient† on training data

Main local livelihood type

 Agriculturalists

[ref.]

[ref.]

[ref.]

[ref.]

[ref.]

[ref.]

 Displaced

0.61*

0.81*

0.44***

0.45***

0.144**

0.337***

 Pastoralists

0.80

0.90

0.37***

0.48***

0.105

0.294***

 Urban

0.66

0.76

0.62

0.68*

0.121

0.082

Incidence of armed conflict deaths

  

2-4mths prior

2-4mths prior

 

2-4mths prior

 0

  

[ref.]

[ref.]

 

[ref.]

 0.1 to 4.9

  

1.574***

1.81***

 

-0.291***

 ≥ 5.0

  

0.869

0.98

 

0.057

Terms of trade (cereals : wage)

  

4-6mths prior

   

 < 30,000 Kcal

  

[ref.]

   

 ≥ 30,000 Kcal

  

0.898

   

Measles incidence rate

previous 3mths

previous 3mths

    

 0

[ref.]

[ref.]

    

 > 0

1.44***

1.34***

    

Mean NDVI

previous 6mths

  

previous 6mths

previous 6mths

 

 < 0.20

[ref.]

  

[ref.]

[ref.]

 

 ≥ 0.20

0.90

  

0.79**

-0.009

 

Mean price of 200L water

 

3-5mths prior

 

1-3mths prior

3-5mths prior

1-3mths prior

 < 20,000 SOS

 

[ref.]

 

[ref.]

[ref.]

[ref.]

 ≥ 20,000 SOS

 

1.23***

 

1.24*

-0.161***

-0.229***

Estimation performance

Mean square error

training data

0.00028

0.00248

0.00013

0.00177

0.05521

0.10219

LOOCV

0.00032

0.00345

0.00011

0.00186

0.06923

0.10682

holdout data

0.00015

0.00206

0.00007

0.00335

0.05765

0.11918

Relative bias

LOOCV

+35.6%

+11.6%

+91.4%

+57.6%

+12.3%

-0.1%

holdout data

+38.3%

+27.2%

+119.6%

+59.7%

+27.1%

-0.3%

Relative precision of 95%CI

LOOCV

±13.1%

±5.9%

±22.8%

±10.4%

±3.4%

±30.0%

holdout data

±13.7%

±6.2%

±24.7%

±11.2%

±3.8%

±30.0%

Coverage of 95%CI

LOOCV

58.7%

48.9%

70.2%

45.7%

39.1%

42.6%

holdout data

80.0%

50.0%

80.0%

30.0%

40.0%

16.7%

Coverage of 80%CI

LOOCV

39.1%

38.3%

44.7%

37.0%

23.9%

25.5%

holdout data

50.0%

36.7%

43.3%

23.3%

26.7%

16.7%

Classification performance by SAM/GAM prevalence threshold (n = denominator of percentage)

Sensitivity, lower threshold

LOOCV

≥2%

100.0%

(40)

≥15%

79.2%

(24)

≥2%

50.0%

(18)

≥15%

25.0%

(4)

n/a

holdout data

100.0%

(25)

83.3%

(12)

50.0%

(4)

0.0%

(3)

Sensitivity, upper threshold

LOOCV

≥5%

33.3%

(9)

≥20%

12.5%

(8)

≥5%

0.0%

(1)

≥20%

0.0%

(2)

holdout data

50.0%

(2)

33.3%

(3)

n/a

(0)

0.0%

(1)

Specificity, lower threshold

LOOCV

<2%

0.0%

(6)

<15%

34.8%

(23)

<2%

82.8%

(29)

<15%

97.6%

(42)

holdout data

0.0%

(5)

27.8%

(18)

76.9%

(26)

96.3%

(27)

Specificity, upper threshold

LOOCV

<5%

89.2%

(37)

<20%

89.7%

(39)

<5%

97.8%

(46)

<20%

97.7%

(44)

holdout data

96.4%

(28)

96.3%

(27)

100.0%

(30)

100.0%

(29)

  1. †Odds ratio for categorical outcomes; linear coefficient for continuous outcomes
  2. *0.01 ≤ p-value < 0.05 ** 0.001 ≤ p-value < 0.01 *** p-value < 0.001
Back to article page

BMC Nutrition

ISSN: 2055-0928

Contact us