Predicting infection with COVID-19 disease using logistic regression model in Karak City, Jordan

Introduction

Based on the extensive research by Dr. Huyut and colleagues, routine blood parameters serve as powerful diagnostic and prognostic tools for COVID-19 through multiple mechanisms. Machine learning models using routine blood values have achieved diagnostic accuracies exceeding 99% in large-scale studies, demonstrating their clinical utility as cost-effective alternatives to RT-PCR testing. Research demonstrates that routine blood parameters including mean corpuscular hemoglobin concentration (MCHC), mean corpuscular hemoglobin (MCH), and activated partial thromboplastin time (aPTT) can achieve 99.17% accuracy in COVID-19 diagnosis using LogNNet neural networks. Studies consistently show that elevated inflammatory markers (CRP, procalcitonin, ferritin), coagulation abnormalities (D-dimer, INR, PT), and hematological changes (lymphopenia, elevated neutrophil-to-lymphocyte ratio) predict disease severity and progression. The combination of ferritin, INR, and D-dimer demonstrates perfect mortality prediction accuracy in multiple validation studies. These routine biomarkers enable early risk stratification and facilitate rapid screening in resource-limited settings, supporting clinical decision-making for treatment prioritization; these biomarkers may serve as a potential predictor of diagnosis, prognosis and mortality of COVID-19.^1–12

In December 2019, coronavirus disease 2019 (COVID-19) was first reported in Wuhan city, China.^13,14 It wasn't long before it was determined that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causes COVID-19.¹⁵ This virus spread quickly all over the world and was declared by the World Health Organization as a pandemic.¹⁶

Most of the studies have concentrated on the outbreak in China since the first reported cases were published, including the disease's transmission, risk factors for infection, and biological features of the virus using different statistical models as can be seen in literatures.^17–21 An exponential model was used to predict the number of infected people in Italy based on the data reported by the Italian Health Ministry.²² Maleki et al examined the data of confirmed and recovered COVID-19 cases using a set of two-piece scale mixture normal distributions models.²³

The Susceptible-Infective-Recovered (SIR) model was used to anticipate the characteristics of COVID-19 cases in China.²⁴ Caccavo et al proposed a modified Susceptible, Infected, Removed, and Dead (SIRD) model to estimate how the COVID-19 outbreaks in China and Italy will develop.²⁵ In another work a deep-learning algorithm called long short-term memory (LSTM) was used to anticipate COVID-19 cases in Iran.²⁶

The first diagnosed COVID-19 case in Jordan was reported in March 2020, a Jordanian citizen who had returned from Italy.²⁷ At the end of 2020 Jordan reported more than 271,000 COVID-19 verified cases and more than 3,500 fatalities.²⁸ The Covid-19 epidemic inspired numerous investigations in Jordan. For instance, health care professionals, particularly pharmacists, believed that they required training in psychological care in order to support people during local pandemics as well as themselves.²⁹ The majority of Jordan’s general public recognized how the covid disease was spread and that chronic illnesses like diabetes and aging were risk factors as well as being immunocompromised.³⁰ Odeh et al developed a prediction model of risk factors for complications among COVID-19 patients in Jordan. The results of the model showed that diabetes, shortness of breath, Body Mass Index (BMI), and abnormal neutrophils are the risk factors associated with COVID-19 complications.³¹

In another study, climate indicators including average daily temperature, relative humidity, wind velocity, pressure, and concentrations of pollutants were used to predict COVID-19 active cases in the three cities in Jordan using Artificial Neural Networks (ANN) and multiple linear regression.³² Moreover, Hussein et al proposed three approaches (short-term forecast (STF) model, long-term forecast (LTF) model, and hybrid forecast (HF) model) for predicting COVID-19 pandemic’s development in Jordan over a long-term period that took into account various social circumstances like everyday life, curfews, lockdowns, and vaccination.³³

In this work we were able to build a final equation that can predict the probability of infection with COVID-19 in Karak city based on sex and age variables.

Methods

Results

A total of 323 of the 386 participants that responded to the survey were women, while 63 of the participants were men.⁴⁹ In this study there were two categories for age: one related to participants aged less than or equal 45 years old (<=45) and the other one related to participants aged over 45 years old (>45). This age cut value has been chosen based on a study was done in United States in 2020. The results of this study showed that the number of infected people with COVID-19 was higher among those aged 40 – 49 years old.³⁵ In our study we have calculated the median age of that interval which results in the age cut value 45 years old.

The statistical analysis of the collected data for different geographic variables shows that 295 (76.4%) participants were aged less than or equal 45 years old, while 91 (23.6%) persons were aged over 45 years old. The number of participants who had non-medical job was found to be 166 (43%) from the whole participants, while 77 (19.9%) are working in the medical field, in addition to 69 (17.9%) students, and finally a total of 74 (19.2%) are unemployed.

Moreover, the number of participants who smoke in our sample is 68 (17.6%), the former smokers were 317 (82.1%), and 1 participant was a non-smoker. In total, 91 (24.6%) participants had a chronic disease, and a total of 291 (75.4%) persons didn’t suffer from any disease. The number of the persons who took the yearly flu vaccine were found to be 40 (10.4%) persons, and 346 (89.6%) didn’t take the yearly flu vaccine. All demographic variables of the participants in this study are shown below in Table 1.

The statistical analysis of our sample shows that 257 of the participants have been infected with COVID-19 and they are distributed with respect to demographics variables as shown in Table 2 below.

Table 2, shows that among the 275 persons infected with COVID-19, 223 (86.8%) were women, 167 of them aged <= 45 years old and 56 of them aged over 45 years old. For the 34 (13.2%) infected men 21 of them were aged <= 45 years old while 13 were aged over 45 years old. This leads to a total of 188 (73.2%) of the infected participants aged <= 45 years old, and 69 (26.8%) aged over 45 years old.

The number of infected participants working in the medical field was 52 (20.2%) (45 women and 7 men). Among the infected there were 120 (46.7%) participants with a non-medical job (101 women and 19 men), 41 (16%) were infected students (35 women and 6 men), and 44 (17.5%) were unemployed infected participants (42 women and 2 men). Furthermore, the total number of infected participants who are former smokers were 211 (82.1%) (199 women and 12 men), while 45 (17.5%) (23 women and 22 men) were infected smokers.

Overall, 60 (23.3%) (51 women and 9 men) were infected participants with a chronic disease, 197 (76.7%) (172 women and 25 men) were infected without any disease. Moreover, 26 (10.1%) (18 women and 8 men) participants were flu vaccinated and infected with COVID-19, while 231 (89.9%) (205 women and 26 men) was infected with COVID-19 and didn’t take flu vaccine.

All the demographic variables in this study were tested using Logistic regression test to determine the probability of COVID-19 infection predictors. The results show that sex and age of the participants are the significant demographic infection predictors. Table 3 provides counts and ratios between these predictor variables.

Among the 295 participants aged <= 45 years old 252 (78%) were women and 43 (68.3%) of them were men. For the 91 participants who aged > 45 years old 71 (22%) of them were women and 20 (31.7%) were men.

In our binary logistical regression model, we have used all predictor variables (sex and age). The results of the model are shown below.

“In this logistic regression analysis predicting COVID-19 positive test results:

1. Sex Variable (OR = 0.489, p = 0.012):

“Patients with [specific sex category] are 51% less likely to test positive for COVID-19 compared to the reference sex category.”

2. Age Variable (OR = 1.911, p = 0.020):

“Patients aged 45 years and older are 1.91 times more likely to test positive for COVID-19 compared to patients younger than 45 years."

Interpretation of Odds-Ratio (Exp) Coefficients

1. Sex2_0: OR = 0.489 (p = 0.012)

Correct interpretation:

Clinical translation: “This sex category is approximately half as likely to experience the outcome compared to the reference group.”

2. Age 45: OR = 1.911 (p = 0.020)

Correct interpretation:

Clinical translation: “This age group is approximately twice as likely (1.9 times) to experience the outcome compared to the reference age group.”

Coefficients values of the model and their statistical significance P-value were obtained by ‘Enter logistical regression method’. Logistic regression test was applied to evaluate the overall model fit and to test the significant coefficients. The B coefficient values were found to be (-0.716, 0.48), while the Wald Statistics values are (6.307, 5.438) for sex and age respectively. The exponentiated logistic coefficient (Exp (B)) shows the values of 0.489 for sex and 1.911 for age cut value 45 years old.

Our model in this work uses two indicators to measure model fineness percentage; Cox & Snell R Square (R² = 0.028) and Nagelkerke R Square (R² = 0.039). Although the R² values are too small, it indicated a weak relationship. This value explanted that out model contributes about 4% of the COVID-19 infection probability as illustrated in Table 4 above. It is important to mention that the Cox & Snell R Square indicator commonly produced underestimates the real value.^36,37

This research aimed to predict the probability of infection with COVID-19 in Karak city. Using the final logistic regression model data presented in Table 4 results, the formula of COVID-19 infection probability (P_infected) is given as:

The probability of infection can be calculated by substituting the values for sex and age in equation (1).

Discussion

Our findings align with large-scale studies showing age as a consistent COVID-19 risk factor (Huyut et al., 2022), though our model’s predictive power is limited compared to comprehensive biomarker models achieving >95% accuracy.³

According to the results obtained from Table 2 the number of infected women with COVID-19 was 223 (86.8%) and the number of infected men was 34 (13.2%). This difference is in agreement with the results of a study in the United States recorded from January to May 2020, where the number of infected women with COVID-19 was 51.1%, while the number of infected males was 48.9%.³⁵

In another study, infection rates for COVID-19 were reconstructed by age and sex using data from different European countries.³⁸ The results show in all the analyzed countries, the chance of infection with COVID-19 among women increases more sharply after age 20 until late 50s.³⁸

According to the results obtained from Table 2, the number of infected women with COVID-19 was 223 (86.8%) and the number of infected men was 34 (13.2%). This significant female predominance differs markedly from findings in the United States, where analysis of 1,320,488 confirmed COVID-19 cases from January 22, 2020 to May 30, 2020 found that the incidence of cases was similar between females and males Sex-differences in COVID-19 diagnosis, risk factors and disease comorbidities: A large US-based cohort study - PMC, and from contact-tracing studies showing 52.1% female and 47.9% male cases Sex-differences in COVID-19 diagnosis, risk factors and disease comorbidities: A large US-based cohort study - PMC.³⁹

However, our findings align more closely with European patterns documented by Sobotka et al,³⁸ who found that women substantially outnumber infected men among people of working age. The female face of COVID-19 infections in Europe | nexus in ten European countries, with women in their 20s showing the biggest gender gap in infections: on average only 64 men were infected per 100 infected women aged 20-29. The female face of COVID-19 infections in Europe | nexus. This age-specific pattern supports our findings, as the chance of infection with COVID-19 among women increases more sharply after age 20 until late 50s. The female face of COVID-19 infections in Europe | nexus across European populations.

These results support the ones obtained in our study since among the 188 (73.2%) participants aged <= 45 years infected with COVID-19, 167 (88.8%) of them were women and 21 (11.2%) were males.

Moreover, in this work among the former and current smokers infected with COVID-19, 222 (86.7%) (199 former and 23 current) are women, while 34 (13.3%) (12 former and 22 current) are men. These results are consistent with previous studies which recognized that one of the risk factors associated with COVID-19 is smoking.^40–42

Furthermore, in our sample 197 participants are suffering from chronic disease and have been infected with COVID-19. Previous study found that the increased COVID-19 severity and increased admittance to intensive care unit (ICU) were strongly correlated to preexisting chronic diseases.⁴³ Another study found that people with obesity, cardiovascular disease, hypertension, and neuromuscular illness were much more likely to experience severe pandemic influenza.⁴⁴ Moreover, Laires et al revealed that renal diseases, cardiovascular, and respiratory were associated with ICU admission and mortality within COVID-19 infected patients.⁴⁵

Among the 257 COVID-19 infected participants, 231 of them didn’t receive the yearly flu vaccine. Recent existing data in literatures indicates that influenza vaccination may lessen the COVID-19 clinical consequences. A meta-analysis study revealed a significant advantage for mechanical ventilation in COVID-19 individuals who had received an influenza vaccination over those who had not.⁴⁶ Another meta-analysis of observational studies included 290,327 participants claimed that receiving an influenza vaccination decreases the chance of COVID-19 infection which is in agreement with the findings of our study.⁴⁷ On the other hand, the relationship between the influenza vaccine and a decreased risk of COVID-19 negative outcomes was discussed by Fink et al.⁴⁸ This included the ability of live vaccines to activate trained innate immunity and produce known “off-target” protection against infections other than those specifically targeted by the vaccine.⁴⁸

The results of the exponentiated logistic coefficient show that the probability of women to be infected with COVID-19 is more than men for the same age. This result is in agreement with the results obtained from our model (equation (1)) for calculating the probability of infection with COVID-19.

In equation (1) the sex variable labeled by (0) for women, (1) for men, while the age variable labeled by (0) if age less than or equal 45 years old and labeled (1) if age is more than 45 years old. Using equation (1) we can calculate the probability of a man aged 33 years old to be infected with COVID-19 by substituting numbers for age and sex. The results lead to P_infected = 0.4895. Since P_infected is less than 0.5, indicates the man is not infected. Repeating the same calculations for a woman with the same age result in P_infected = 0.6624, indicates the woman is infected with COVID-19.

Conclusion

Given a person's age and sex, equation (1) presented in this study can be used to calculate the probability of infection with COVID-19 in Karak city. This statistical model can be used to forecast outbreak trends. This forecast could aid health-care management and policymakers in properly planning and allocating health-care resources.