Developing Prediction Models for COVID-19 Outcomes: A Valuable Tool for Resource-Limited Hospitals

Figures & data

Table 1 Distribution of Demographic, Clinical Characteristics, and Comorbidities of Patients Across the Different Severity Forms Subgroups

Variables	Result	Mean or N (%)
		Total	Mild	Moderate	Severe	p-value
		N=483 (100%)	N=190 (39.33%)	N=175 (36.23%)	N=118 (24.43%)
Age		61.19	56.85	61.44	67.83	<0.001^a
Sex	Male	271(56.11)	87(45.79)	105(60)	79(66.95)	<0.001^b
	Female	212(43.89)	103(54.21)	70(40)	39(33.05)
CVD		308(63.77)	108(56.84)	110(62.86)	90(76.27)	0.002^b
CPD		51(10.56)	23(12.11)	14(8.00)	14(11.86)	0.385^b
Obesity	I	37(7.66)	9(4.74)	18(10.29)	10(8.47)	0.128^b
	II	30(6.21)	7(3.68)	13(7.43)	10(8.47)	0.162^b
	III	11(2.28)	3(1.58)	3(1.71)	5(4.24)	0.288^c
	Severe	11(2.28)	5(2.63)	4(2.29)	2(1.69)	0.928^c
	All forms	89(18.42)	24(12.63)	38(21.71)	27(22.88)	0.029^b
Diabetes mellitus	Type 1	35(7.25)	13(6.84)	13(7.43)	9(7.63)	0.960^b
	Type 2	76(15.73)	23(12.11)	31(17.71)	22(18.64)	0.206^b
	Both types	111(22.98)	36(18.94)	44(25.14)	31(26.27)	0.231^b
Hepatopathy	Hepatitis B	5(1.04)	2(1.05)	3(1.71)	0(0.00)	0.375^c
	Hepatitis C	8(1.66)	3(1.58)	2(1.14)	3(2.54)	0.603^c
	Cirrhosis	9(1.86)	4(2.11)	1(0.57)	4(3.39)	0.201^c
	All pathologies	22(4.55)	9(4.73)	6(3.42)	7(5.93)	0.594^c
Renal disease		37(7.66)	9(4.74)	8(4.57)	20(16.95)	<0.001^a
Neurologic disorder		37(7.66)	9(4.74)	15(8.57)	13(11.02)	0.111^b
Immunodeficiency		2(0.41)	1(0.53)	0(0.00)	1(0.85)	0.714^c
Autoimmune disease		10(2.07)	5(2.63)	4(2.29)	1(0.85)	0.658^c
Cancer		19(3.93)	9(4.74)	4(2.29)	6(5.08)	0.365^c

Notes: ^aKruskal–Wallis H-test p-value; ^bChi-squared test p-value; ^cFisher’s exact test (the Freeman-Halton extension) p‑value.

Table 2 Comparison of Paraclinical Factors Values Across the Different Severity Forms Subgroups

Variable	Mean (Min-Max) or N (%)
	Total (N=462)	Mild (N=189)	Moderate (N=174)	Severe (N=99)	p-value
WBC	8.2689 (1.35–48.08)	7.065 (1.73–23.55)	8.116 (1.35–26.33)	10.834 (2.11–48.08)	<0.001
Increased (Leucocytosis)	121(26,19)	33(17.46)	44(25.29)	44(44.44)	<0.001
Normal	341(73,81)	156(82.54)	130(74.71)	55(55.56)
ANC	6.4727 (0.38–37.84)	5.1528 (0.38–21.96)	6.4624 (0.62–24.57)	9.0107 (1.43–37.84)	<0.001
Increased (Neutrophilia)	171(37.01)	46(24.33)	63(36.2)	62(62.62)	<0.001
Normal	421(62.99)	143(75.67)	111(63.8)	37(37.38)
	Total (N=455)	Mild (N=187)	Moderate (N=171)	Severe (N=97)
CRP value	74.7093 (0.04–368.81)	41.9270 (0.04–272.86)	87.7086 (0.45–368.81)	114.9919 (1.63–365.96)	<0.001
Increased	412(90.54)	153(81.81)	165(96.49)	94(96.9)	<0.001
Normal	43(9.46)	34(18.19)	6(3.51)	3(3.1)

Table 3 Disease Outcome in the Study Group and Severity Forms Subgroups

Outcome	N(%)
	Total (N=483)	Mild (N=190)	Moderate (N=175)	Severe (N=118)	p-value
Unfavourable	99(20.5)	10(5.27)	17(9.71)	72(61.01)	<0.001
Favourable	384(79.5)	180(94.73)	158(90.29)	46(38.99)

Table 4 Association Between Various Demographic Characteristics, Disease Severity Form, Comorbidities, and Unfavourable Outcome in the Study Group, Based on Spearman Correlation and Univariate Logistic Regression Analysis

Risk Factor	Unfavourable Outcome
	Spearman Correlation		Univariate Logistic Regression
	Rho	p-value	cOR (95% CI)	p-value
Demographic characteristics
Age	0.3852	<0.001	1.092(1.069–1.118)	<0.001
Male sex	0.0150	0.741	1.078(0.691–1.691)	0.741
Disease severity
Mild form	−0.3038	<0.001	0.127(0.060–0.241)	<0.001
Moderate form	−0.2013	<0.001	0.296(0.164–0.507)	<0.001
Severe form	0.5707	<0.001	19.594(11.576–34.103)	<0.001
Comorbidities
CVD	0.2013	<0.001	3.372(1.969–6.081)	<0.001
CPD	−0.0576	0.206	0.587(0.235–1.270)	0.209
Obesity	0.0364	0.423	1.251(0.710–2.136)	0.423
Diabetes mellitus	0.1005	0.027	1.732(1.051–2.814)	0.028
Hepatopathy	0.0366	0.421	1.483(0.520–3.718)	0.423
Renal disease	0.2008	<0.001	4.269(2.133–8.527)	<0.001
Neurologic disorder	0.2394	<0.001	5.465(2.742–11.019)	<0.001
Immunodeficiency	0.0471	0.301	3.908(0.153–99.416)	0.336
Autoimmune disease	−0.0378	0.407	0.425(0.022–2.302)	0.419
Cancer	0.0027	0.951	1.035(0.290–2.930)	0.951

Abbreviations: Rho, Spearman’s rank correlation coefficient; cOR, crude odds ratio for logistic univariate regression; 95% CI, 95% confidence interval.

Table 5 Correlation Between Paraclinical Factors and Unfavourable Outcome in the Study Group, Based on Spearman Correlation and Univariate Logistic Regression Analysis

Predictive Factor	Unfavourable Outcome
	Spearman Correlation		Univariate Logistic Regression
	Rho	p-value	cOR (95% CI)	p-value
WBC	0.2434	<0.001	1.103(1.053–1.159)	<0.001
ANC	0.2699	<0.001	1.153(1.094–1.218)	<0.001
CRP	0.2435	<0.001	1.007(1.004–1.009)	<0.001

Abbreviations: Rho, Spearman’s rank correlation coefficient; cOR, crude odds ratio for logistic univariate regression; 95% CI, 95% confidence interval.

Table 6 Correlation Between Selected Clinical Risk Factors and Unfavourable Outcome in the Study Group Using Multivariate Logistic Regression

Variables	aOR (95% CI)	p-value
Age	1.067(1.037–1.099)	<0.001***
Severe form	1.811(1.002–34.080)	<0.001***
CVD	1.506(7.282–3.206)	0.275
Renal disease	1.212(4.822–3.008)	0.678
Neurological disorders	2.914(1.106–7.647)	0.029*

Notes: “***”: p-value ≤ 0.001; “*”: 0.01 < p-value ≤ 0.05; “ ”: 0.1 < p-value ≤ 1 (not significant).

Abbreviations: aOR, adjusted odds ratio for logistic multivariate regression; 95% CI, 95% confidence interval.

Table 7 Correlation Between Paraclinical Risk Factors and Unfavourable Outcome in the Study Group Using Multivariate Logistic Regression

Variables	aOR (95% CI)	p-value
WBC	0.741(0.495–1.013)	0.118
ANC	1.533(1.105–2.332)	0.031*
CRP	1.004(1.001–1.008)	0.002**

Notes: “**”: 0.001 < p-value ≤ 0.01; “*”: 0.01 < p-value ≤ 0.05; “ ”: 0.1 < p-value ≤ 1 (not significant).

Abbreviations: aOR, adjusted odds ratio for logistic multivariate regression; 95% CI, 95% confidence interval.

Figure 1 Receiver operating characteristic (ROC) curves illustrating the performance of proposed predictive models in distinguishing high-risk and low-risk patients within the study group dataset.

Figure 2 Receiver operating characteristic (ROC) curves illustrating the performance of proposed predictive models in distinguishing high-risk and low-risk patients within the extended group dataset.