Predicting COVID-19 Re-Positive Cases in Malnourished Older Adults: A Clinical Model Development and Validation

Introduction

COVID-19 infection has the characteristics of high mutation rate and a high transmissibility, ¹ as of July 9, 2023, more than 767 million confirmed cases and more than 6.9 million deaths have been reported worldwide.² Currently, China has entered a period of deep aging, with a large base and fast speed of aging population, ^3,4 older adults are vulnerable groups during COVID-19 epidemic due to complex underlying diseases, atypical clinical symptoms, and multiple complications.^5,6 Studies have found that COVID-19 infection is a high-risk group for nutritional risk and malnutrition, with rates of nutritional risk and malnutrition as high as 77% and 50%, respectively.^7–10 A study in Wuhan, China, showed that among patients infected with COVID-19 aged over 65 years, the rates of nutritional risk and malnutrition were 27.5% and 52.7%, respectively.¹¹ In addition, malnutrition can activate related inflammatory pathways, thereby damaging the human immune system. For older adults and critically ill patients with multiple underlying diseases, early initiation of enteral and parenteral nutrition support can effectively protect the intestinal mucosal barrier and immune function, reduce the incidence and mortality of infectious complications.^12–14 Therefore, timely and standardized nutritional support can effectively improve the nutritional status, immune function, and clinical outcomes of COVID-19 infection, ¹⁵ promote recovery and reduce the risk of recurrence, ^16,17 improve patient survival and quality of life.Therefore, how to quickly and accurately foresee the risk factors of COVID-19 “re-positive” in malnourished older adults is a clinical problem that needs to be solved urgently.

In recent years, clinical prediction models based on machine learning have helped to further understand the important determinants of disease risk, more accurately assess the condition and prognosis, and also facilitate communication and cooperation between doctors and patients by screening high-risk factors of disease as predictors and building relevant models to predict the probability of outcome occurrence.^18–21 This has an important role in the three-level prevention system of disease. This study retrospectively analyzed the clinical data of 347 malnourished older adults, included relevant influencing factors affecting COVID-19 re-positive in malnourished older adults and constructed a clinical prediction model, aiming to predict the possible risk factors affecting COVID-19 re-positive, so as to intervene early, assist clinical decision-making, optimize clinical management, and improve patient prognosis.

Materials and Methods

Study Subjects and Inclusion and Exclusion Criteria

Older adults with malnutrition in the geriatric department of the Affiliated Hospital of Chengdu University of Traditional Chinese Medicine were retrospectively collected from January to May 2023. According to the number of COVID-19 infections, they were divided into “non-re-positive” and “re-positive” groups, and divided into training and testing sets at a ratio of 7:3.

Inclusion Criteria

(1) age ≥ 65 years; (2) meeting the diagnostic criteria for malnutrition; (3) “non-re-positive” patients who had not been infected, had been infected once, or were currently infected with COVID-19 for the first time; (4) “re-positive” patients who met the diagnostic criteria for COVID-19 “re-positive”; (5) complete relevant clinical data and informed consent.

Exclusion Criteria

(1) incomplete relevant clinical data; (2) patients with dementia, mental illness, or who are unable to cooperate with data collection (Figure 1).

Figure 1 Patient Flow Diagram.

Diagnostic Criteria

Older Adults Malnutrition Diagnostic Criteria

The Geriatric Nutritional Risk Index (GNRI) proposed by the American Society for Clinical Nutrition²² was used: GNRI=1.489*serum ALB(g/dL)+41.7*(actual weight/ideal weight). Ideal weight was calculated by the Lorentz formula: ideal weight=22*height(m)*height(m), If the actual weight > ideal weight, then the actual weight/ideal weight was taken as 1; if the actual weight < ideal weight, then the actual ratio was used for calculation. Scoring criteria: According to the value of GNRI, 92≤GNRI<98, 82≤GNRI<92, and GNRI<82 corresponded to mild malnutrition, moderate malnutrition, and severe malnutrition, respectively.

COVID-19 Diagnostic Criteria

Coronavirus Disease 2019 (COVID-19) Treatment Guidelines²³ can be referred to. COVID-19 “re-positive” diagnostic criteria: (1) patient had previously been infected with COVID-19 and tested positive for nucleic acid, and then tested negative for nucleic acid after improvement, and then tested positive for nucleic acid again; (2) false positives caused by specimen collection and testing were excluded.

Procedure

The general clinical data and laboratory indicators of the patients in this study were obtained by reviewing the electronic record system of the Affiliated Hospital of Chengdu University of Traditional Chinese Medicine and by asking the patients themselves or their families, and the specific items included age, body mass index (BMI), gender, GNRI, cerebral infarction, renal failure, tumor, pneumonia, chronic obstructive pulmonary disease, type 2 diabetes, coronary disease, hypertension, skin ulcer, history of drinking, history of smoking, vaccination, white blood cell (WBC), neutrophil granulocyte, lymphocyte, c-reactive protein (CRP), d-dimer, hemoglobin (Hb), serum creatinine (Scr), albumin(ALB), alanine aminotransferase (ALT), aspartate aminotransferase (AST);The measurement of the patients’ nutritional status assessment was mainly referred to as part of the new version of the Mini Nutritional Assessment(MNA^R)²⁴ and made into a questionnaire form, which was measured, evaluated, and filled in by the staff who had received standardized training on the patients.The items included mid-arm circumference, gastrocnemius circumference, protein intake, intake of vegetables and fruits, diet status in recent 3 months, activity situation, and neurological and psychological problems.

Statistical Analysis

Data on gender, age, height, weight, BMI, underlying diseases, vaccination status, diet, and laboratory results were collected for all study subjects. According to the COVID-19 infection status of the patients, they were divided into “non-re-positive” and “re-positive” groups. Statistical analysis was performed using R-4.2.2 and SPSS 26.0 software. If the measurement data followed a normal distribution, they were expressed as mean±standard deviation () and independent samples t-test was used; if they followed a non-normal distribution, they were expressed as median (quartile) (M(Qn)) and Mann–Whitney U-test was used; count data were expressed as number (percentage) (n(%)) and χ² test was used. LASSO regression was used to reduce the dimensionality of the included variables and screen out the influencing factors of COVID-19 “re-positive” in malnourished older adults; further, a clinical prediction model was established using multivariate logistic regression and a nomogram was drawn;Hosmer-Lemeshow test was used to evaluate the model fit; and χ² statistic and P value were calculated; a calibration curve was drawn; an AUC was used to judge the discrimination of the model; and DCA and CIC were used to evaluate the validity and clinical impact of the model. P<0.05 indicated that the difference was statistically significant.

Results

Comparison of General Data Between Training and Testing Sets

A total of 347 malnourished older adults were finally included, and divided into training and testing sets at a ratio of 7:3, with 243 and 104 cases, respectively. There were no statistically significant differences between the training and testing sets in age, gender, BMI, mid-arm circumference, calf circumference, smoking history, drinking history, vaccination status, activity status, psychological status, skin condition, dietary status in the past 3 months, vegetable and fruit intake, protein intake, hypertension, type 2 diabetes mellitus, coronary heart disease, cerebral infarction, chronic obstructive pulmonary disease, pneumonia, renal failure, tumor, WBC, neutrophil granulocyte, lymphocyte, CRP, Hb, d-dimer, ALB, ALT, AST, Scr (P>0.05), (Table 1).

Table 1 Comparison of Clinical Information Between Training Set and Validation Set((n(%)) or Cases/%)

LASSO Regression Screening of Predictive Factors for COVID-19

“re-positive” in malnourished older adults LASSO regression was used to reduce the dimensionality of the included variables. A total of 10-fold cross-validation was performed to analyze the correlation among the variables. The LASSO coefficient curve of the included variables was obtained as shown in Figure 2A. When the distance from the mean square error was one standard error (λ1se), that is, λ1se=0.054 and log(λ)=−2.917, the model was optimal. A total of 9 non-zero coefficient predictive variables were screened out as influencing factors for COVID-19 “re-positive”, as shown in Figure 2B. They were as follows: age, vaccination status, vegetable and fruit intake status, protein intake amount, renal failure, lymphocyte, Hb, d-dimer and AST.

Figure 2 Plots for LASSO regression model. (A) 10-fold cross-validation plot for the penalty term. (B) A LASSO coefficient profiles plot of the 33 texture features was produced against the log (lambda) sequence.

Logistic Regression Analysis of Predictive Factors for COVID-19 “Re-Positive” in Malnourished Older Adults

Whether COVID-19 infection was “re-positive” or not was used as the dependent variable. The 9 predictive variables screened by LASSO regression analysis and GNRI which was closely related to this study were used as independent variables.^22,25 Multivariate logistic regression analysis was performed to calculate the odds ratio (OR) and 95% confidence interval (CI). The size of the confidence interval depends on the sample size and the standard deviation of the study group.²⁶ If the sample size is large, this leads to “greater confidence” and narrower confidence intervals; if the confidence interval is wide, this may mean that the sample is small; and if the dispersion is high, there is less certainty in the conclusions and wider confidence intervals.²⁷ A positive correlation between exposure and outcome implies an OR > 1.0, and a negative correlation implies an OR < 1.0.²⁸ Multifactorial logistic regression analyses in this study showed that age (OR 1.041; 95% CI 0.991–1.093), GNRI (OR 0.738; 95% CI 0.381–1.430), vaccination (OR 0.816; 95% CI 0.491–1.355), Intake of vegetables and fruits (OR 0.564; 95% CI 0.309 −1.028), protein intake (OR 0.588; 95% CI 0.353–0.982), renal failure (OR 9.299; 95% CI 2.580–33.516), lymphocyte (OR 0.121; 95% CI 0.060–0.242), Hb (OR 0.990; 95% CI 0.973–1.008); d-dimer (OR 1.236; 95% CI 0.901–1.696); AST (OR 1.016; 95% CI 1.005–1.028), where the CI for Renal failure was significantly larger, which may be related to the limited sample size and the inability to fully correct for bias and confounding.In addition, in this study, when the regression coefficients in the regression analysis were positive and the OR value was >1, the factors were determined to be risk factors affecting the outcome; conversely, they were protective factors, ie, high GRNI, higher number of doses of vaccination, high intake of vegetables and fruits, high intake of proteins, high lymphocytes, and high Hb were protective factors, and high age, renal failure, high d-dimer, and high AST were risk factors, as shown in Table 2. In addition, vegetable and fruit intake, protein intake, lymphocytes, renal failure, and AST were independent influences (P<0.05). Meanwhile, the Hosmer-Lemeshow test yielded χ² = 5.916, P= 0.657 (P>0.05), indicating that the predictive model fit was good.

Table 2 Multifactorial Logistic Regression of Predictive Factors for “Re-Positive” COVID-19 Infection in Malnourished Older Adults

Establishment of a Clinical Prediction Model

A clinical prediction model nomogram was constructed using the 10 variables included in the logistic regression model as predictive factors and whether COVID-19 infection was “re-positive” or not in malnourished older adults as clinical outcome. The length of each variable segment in the nomogram was positively correlated with its influence on the clinical outcome. The total score was the sum of the scores of each variable. The scale value corresponding to the total score was the risk probability of COVID-19 infection “re-positive” in malnourished older adults; see Figure 3.

Figure 3 The nomogram based on the multivariable regression model. Abbreviations: GNRI, Geriatric Nutritional Risk Index;Hb, hemoglobin; AST, aspartate aminotransferase.

Validation of a Clinical Prediction Model

The above model was validated using data from 104 cases in the testing set. The predictive performance of the prediction model was evaluated from three aspects: discrimination, calibration and clinical impact. The AUC of the training set was 0.881 (95% CI: 0.839–0.923), see Figure 4A, and that of the validation set was 0.896 (95% CI: 0.835–0.956), see Figure 4B, indicating that the prediction model had high discrimination and good discrimination ability. At the same time, the calibration curve suggested that when the predicted probability was 15%-70%, the model prediction curve was close to the actual observation curve, indicating that the prediction model had good calibration ability; see Figure 5A and B. According to DCA analysis, when the threshold probability >8%, using this prediction model to predict whether COVID-19 infection is a “re-positive” risk in malnourished older adults is more beneficial than implementing an intervention plan for all patients. The net benefit of the prediction model was significantly higher than that of all or no intervention; see Figure 6A and B. Based on DCA, CIC was further drawn to evaluate the clinical impact of the model, showing the estimated number of positive cases and the actual number of cases at different risk thresholds. The results suggested that when the risk threshold was >80%, the positive estimate value was close to the actual number of cases. See Figure 7A and B.

Figure 4 ROC curves. (A) Training set. (B) Validation set.

Figure 5 Calibration curves. (A)Training set. (B) Validation set.

Figure 6 Decision curve analysis. (A)Training set. (B) Validation set.

Figure 7 Clinical impact curve analysis.(A)Training set. (B) Validation set.

Discussion

Since January 8, 2023, China has implemented “Class B management” for COVID-19 infection, and the focus of prevention and control has shifted to “maintaining health and preventing severe cases.” The incubation period of COVID-19 variants has shortened, and the pathogenicity has shown a significant downward trend, but related studies have shown that the risk of developing severe or fatal cases is still high for older adults, ²⁹ another study found that malnourished older adults are more likely to develop severe cases, ^30,31 leading to an increase in severe and mortality rates.^32,33 In view of the prominent role of nutritional support in preventing and slowing down the development of infection, it is of great guiding significance for how to carry out the next step of nutritional support and predict the disease progression and prognosis of malnourished older adults to foresee the risk factors of COVID-19 “re-positive” in malnourished older adults as early as possible. At present, there are still few studies on COVID-19 “re-positive” in malnourished older adults at home and abroad. This study retrospectively analyzed the influencing factors of COVID-19 “re-positive” in malnourished older adults, screened risk factors by dimensionality reduction, constructed and validated models, aiming to provide a basis for accurately identifying high-risk population patients and provide reference for early intervention and medical decision-making.

A total of 347 valid cases were included in this study, with 243 cases in the training set and 104 cases in the testing set. The 9 predictive variables screened by LASSO regression were age, vaccination status, vegetable and fruit intake status, protein intake amount, renal failure, lymphocyte, Hb, d-dimer, AST and GNRI which were closely related to this study as factors affecting the outcome. Through multivariate logistic regression analysis, it was preliminary judged that high GNRI, more doses of vaccination, high vegetable and fruit intake amount, high protein intake amount, high lymphocyte and high Hb were protective factors; age, renal failure, high d-dimer and high AST were risk factors. Among them, protein intake amount, lymphocyte, renal failure and AST were independent influencing factors. A clinical prediction model for COVID-19 “re-positive” in malnourished older adults was constructed. Hosmer-Lemeshow test yielded χ² =5.916, P=0.657, indicating that the prediction model had a good fit. Both the training set and the testing set showed that the prediction diagnosis was consistent with the actual diagnosis. The AUC of the prediction model in the training set was 0.881 and that in the testing set was 0.896, indicating that the prediction model had good accuracy. At the same time, DCA and CIC also suggested that the prediction model had high clinical validity and impact. The clinical prediction model for COVID-19 “re-positive” in older adults malnourished patients showed that controlling underlying diseases and balancing diet can help reduce the risk of COVID-19 “re-positive” in older adults.

Protein intake amount, renal failure, lymphocyte and AST were independent influencing factors for COVID-19 “re-positive” in malnourished older adults. Previous studies have shown that energy consumption and protein breakdown are accelerated in COVID-19 infection patients, ³⁴ resulting in a reduction of available protein in the body, leading to a decrease in the number of functional active immunoglobulins and intestinal-associated lymphoid tissue, hindering the role of intestinal mucosal defense against infection;^35,36 another study³⁷ found that patients with comorbidities were more likely to have re-positive after discharge, compared with those with normal liver or kidney function, re-positive patients with liver or kidney function impairment had a 2.32-fold higher risk of developing severe cases, ^38–40 and the tendency of severity and mortality risk increased significantly. Therefore, adequate intake of high-quality protein is essential for antibody production, ⁴¹ while for patients with liver or kidney dysfunction, protein intake should be strictly and standardized controlled.⁴² Lymphocyte count can predict disease severity, hospitalization time and prognosis of COVID-19 patients, ^43–45 and several studies have shown^46,47 that lymphocyte count can be used as a risk factor for re-positive rate of COVID-19 patients, and has certain clinical value for predicting recurrence, which is consistent with this study’s lymphocyte as an independent influencing factor and protective factor for “re-positive”. Age, renal failure, high d-dimer and high AST were risk factors for “re-positive”. Some studies^48,49 have shown that elevated levels of ALT, AST, erythrocyte sedimentation rate, d-dimer and decreased neutrophil count are risk factors for re-positive. Previous studies have found that whether COVID-19 infection is “re-positive” or not is related to the age of the patient, ⁴⁶ and due to the poor health status and low immunity of older adults, the virus clearance rate and tolerance in the body are relatively weak, which makes it easier to cause re-positive.⁵⁰ High GNRI, more doses of vaccination, high vegetable and fruit intake amount, high protein intake amount, high lymphocyte and high Hb were protective factors for “re-positive”. GNRI is the geriatric nutritional risk index, ²² which integrates three indicators of serum albumin level, actual weight and ideal weight, and GNRI score is proportional to the nutritional status of the patient, and can more fully reflect the nutritional status of the patient’s body.^51,52 Related studies have shown⁵³ that GNRI assessment may provide valuable information for predicting the prognosis of COVID-19 pneumonia in older adults. Several previous studies have attempted to compare the GNRI with other standardized indices such as the MNA and the GNRI, and have found the GNRI to be valid for use in older adults, but mostly in combination with other indices.⁵⁴ In terms of ongoing exploratory studies on the validity of the GNRI in Cairo and Egypt, it was found that the GNRI is more suitable for the classification of nutritional status and the identification of nutrition-related complications in hospitalized older adults than the MNA.⁵⁵ In terms of using the GNRI to assess the nutritional status of older adults with specific diseases it was found that the GNRI predicts mortality in hemodialysis patients, that it is strongly related to exercise tolerance, that it can be used as a nutritional assessment scale for older patients with chronic obstructive pulmonary disease, that it predicts survival time in older patients with squamous cell carcinoma of the esophagus, and that the GNRI provides a reliable assessment of patients with a potential need for nutritional support, . Especially for older adults with dementia and aphasia.^55–58 In addition, the GNRI is time-consuming, easy to use, and requires minimal involvement to help clinical staff (especially dietitians) diagnose malnourished individuals, which is one of the reasons why the GNRI has been accepted as a tool for assessing the nutritional status of hospitalized older adults.Vaccines are the “ultimate weapon” to end infectious diseases, and related studies have shown that vaccinating older adults can effectively reduce the risk of developing severe or critical cases or even death after COVID-19 infection, ^59,60 and enhancing immunity can help further improve the protective effect.⁶¹ A series of studies in Malaysia found⁶² that the vaccine efficacy in intensive care admissions for older adults was significantly reduced. If this study is confirmed in other studies, increasing supplemental protection for severe disease and death (especially for older adults) before the expected wave of infection, strengthening vaccination is still a key tool to reduce the burden and mortality of COVID-19 on the health care system.⁶³ Many studies have shown that eating more fruits, vegetables and whole grain foods can reduce the risk and incidence of COVID-19 pneumonia.^64–66

Compared with previous studies, we have the following advantages. First, we based our study on a malnourished older adult population and then explored the risk factors associated with the COVID-19 re-positive, which has rarely been reported in previous studies. Relevant studies have shown that advanced age and malnutrition are all risk factors for susceptibility to COVID-19, ²⁹ so exploring the risk factors associated with COVID-19 re-positive in malnourished older adults could help minimize the occurrence of adverse prognostic events. Second, we may be the first to develop a model to predict COVID-19 re-positive based on malnourished older adults, which is important for early diagnosis and prevention of the disease. Of course, our study has some limitations. First, as a cross-sectional study, our sample size was limited, which to some extent also affected the interpretation of the results of the confidence intervals of the associated factors. Second, the risk prediction model was only validated with an internal dataset, whereas validation with an external dataset is necessary. Therefore, our later research will expand the sample size, and we will also collaborate with several centers to obtain external data to validate the model.

Conclusion

In summary, this study exploratorily constructed a clinical prediction model for COVID-19 “re-positive” in malnourished older adults, and intuitively showed the risk probability of COVID-19 “re-positive” by drawing a nomogram. The analysis results showed that the model had good predictive performance, which could help clinical workers accurately assess patient prognosis. At the same time, the model suggested that controlling underlying diseases and balancing diet could help reduce the risk of COVID-19 “re-positive”, which could provide a basis for clinical decision-making and improve patient prognosis; in addition, accurate prevention of COVID-19 “re-positive” could enable patients to achieve health and economic benefits. However, as time changes, the risk factors of disease, unmeasured risk factors, treatment measures and so on are changing dynamically, which may lead to a decline in the standardization and performance of the model; in addition, there are many tools to evaluate the nutritional status of older adults, this study is a preliminary exploratory study, and the research methods are relatively single. Therefore, in future studies, we can include as many relevant influencing factors as possible, expand the sample size, conduct multi-center, multi-tool, multi-dimensional comprehensive research, to reduce information bias, and constantly improve and update the model.