Guidelines for Triage of COVID-19 Patients Presenting with Multi-systemic Symptoms

Abstract

Background and Objectives:

This paper describes how multi-systemic symptoms, both respiratory and non-respiratory, can be used to differentiate COVID-19 from other diseases at the point of patient triage in the community. The paper also shows how combinations of symptoms could be used to predict the probability of a patient having COVID-19.

Materials and Methods:

We first used a scoping literature review to identify symptoms of COVID-19 reported during the first year of the global pandemic. We then surveyed individuals with reported symptoms and recent RT-PCR test results to assess the accuracy of diagnosing COVID-19 from reported symptoms. The scoping literature review which included 81 scientific articles published by February 2021 identified 7 respiratory, 9 neurological, 4 gastrointestinal, 4 inflammatory, and 5 general symptoms associated with COVID-19 diagnosis. The likelihood ratio associated with each symptom was estimated from sensitivity and specificity of symptoms reported in the literature. 483 individuals were then surveyed to validate the accuracy of predicting COVID-19 diagnosis based on patient symptoms using the likelihood ratios calculated from the literature review. Survey results were weighted to reflect age, gender, and race of the U.S. population. The accuracy of predicting COVID-19 diagnosis from patient-reported symptoms was assessed using Area under the Receiver Operating Curve (AROC).

Results:

In the community, cough, sore throat, runny nose, dyspnea, and hypoxia, by themselves, were not good predictors of COVID-19 diagnosis. A combination of cough and fever was also a poor predictor of COVID-19 diagnosis (AROC = 0.56). The accuracy of diagnosing COVID-19 based on symptoms was the highest when individuals presented with symptoms from different body systems (AROC of 0.74 to 0.81); the lowest accuracy was when individuals presented with only respiratory symptoms (AROC=0.48).

Conclusions:

There are no simple rules that clinicians can use to diagnose COVID-19 in the community when diagnostic tests are not available or untimely. However, triage of patients to appropriate care and treatment can be improved by reviewing the combinations of certain types of symptoms across body systems.

Background

COVID-19 is a systemic disease with respiratory, neurological, gastrointestinal, and other symptom manifestations. ^[1–4] A thorough symptom screening, that goes beyond respiratory symptoms (e.g., cough and fever), can improve clinicians’ triage decisions. ^[5–12] Presently, primary care providers continue to screen patients via phone and online screening questionnaires prior to in-person visits to medical offices. Many clinics continue to screen patients, repeatedly, at the entrance to their facilities. These screening procedures rely on a limited list of symptoms, per the guidelines from the U.S. Centers for Disease Control and Prevention (CDC), and often result in triaging patients who present with symptoms similar to COVID-19 to the COVID-19-restricted areas. Incorrectly identifying COVID-19 in healthy individuals bears a number of adverse consequences. First, it puts healthy individuals at risk of actually acquiring the disease as a result of being misidentified and placed in same quarantine wards with actual COVID-19 patients. Second, it prevents healthy individuals from access to timely medical care, which delays proper diagnosis and treatment, at times lifesaving. Third, misidentification of healthy individuals for highly infectious ones puts further pressure on the health system by misallocating and wasting valuable resources.

This paper reviews the literature for multi-systemic symptoms reported by COVID-19 patients and develops a model to predict the probability of COVID-19 diagnosis from both respiratory and non-respiratory symptoms. We report the accuracy of symptom screening for COVID-19 based on the likelihood ratios estimated from symptom prevalence among COVID-19 and non-COVID-19 patients. A comprehensive and accurate symptom screening can 1) improve quarantine protocols for return to work or school, 2) improve timely access to much needed medical care and 3) make better use of already strained resources within healthcare ecosystem. Furthermore, improved screening protocols can better guide surveillance initiatives, the interpretations of at-home rapid tests, and clinicians’ triage decisions.

Materials and Methods

Data were obtained from two sources: a scoping literature review and a survey. In the scoping literature review, we first identified all symptoms reported in the literature that were associated with COVID-19. We then identified the same symptoms that were reported in the literature that were not associated with COVID-19. Having identified the prevalence of symptoms in COVID-19 and non-COVID-19 patients, we generated likelihood ratios (LRs) of symptoms as predictors of COVID-19. This determined the diagnostic value associated with each symptom. We then tested those LRs, applying them to a new cohort of survey participants to learn a) if they could predict the odds of a symptomatic individual having COVID and b) the accuracy of diagnosing patients based on their symptoms.

We searched PubMed, between October 2020 and February 2021, for studies reporting all known symptoms of COVID-19. This study also required a purposeful literature search to find most recent publications that provided the prevalence of symptoms associated with COVID-19 known in non-COVID-19 patients, prior to 2019. Thus, the literature review identified most recent articles on a variety of diagnoses and was organized based on the need to calculate the likelihood ratios for each of the known symptom associated with COVID-19. The search for symptoms was not restricted by population, age, comorbidities, or viral strains. Instead, we took the most recent related publications with the purpose of estimating the likelihood ratios. A total of 81 studies were identified. Most studies included in the review were from years 2020–2021 with the exception of one publication from 2009 which was used to obtain the prevalence of seizures prior to the beginning of the pandemic. We searched for “COVID-19” and “symptoms”. Once all known symptoms of COVID-19 to date were identified in the literature, we searched for the studies that reported the prevalence of those symptoms prior to the emergence of COVID-19. The symptoms identified from the reviewed studies were classified into five categories: respiratory, neurological, gastrointestinal, inflammatory, and general. The following formulas (see below) were used to estimate the likelihood ratios associated with the presence or absence of a symptom; from either the reported sensitivity or specificity of the symptom or the reported prevalence of the symptom in COVID-19 patients. Sensitivity defines the percent of individuals screened as positive for the presence of SARS-CoV-2 infection, among those who actually have the infection. Specificity defines the percent of individuals who were screened negative for SARS-CoV-2 infection, among those who do not have the infection.

$${LR}^{+}=\frac{Sensitivity}{1-Specificity}=\frac{Prevalence\mathrm{ }of\mathrm{ }Symptom\mathrm{ }among\mathrm{ }{COVID19}^{+}}{Prevalence\mathrm{ }of\mathrm{ }Symptom\mathrm{ }among\mathrm{ }{COVID19}^{-}}$$

$${LR}^{-}=\frac{1-Sensitivity}{Specificity}=\frac{1-Prevalence\mathrm{ }of\mathrm{ }Symptom\mathrm{ }among\mathrm{ }{COVID19}^{+}}{1-Prevalence\mathrm{ }of\mathrm{ }Symptom\mathrm{ }among\mathrm{ }{COVID19}^{-}}$$

In above formula, LR⁺ indicates the likelihood ratio associated with the presence of the symptom and LR⁻ indicates the same for the absence of the symptom. The independent form of Bayes formula was used to examine the impact of multiple symptoms on odds of having COVID-19 diagnosis:

$$Posterior\mathrm{ }Odds=Prior\mathrm{ }Odds\mathrm{ }\times {\prod }_S{LR}_s$$

To test the accuracy of the model, 483 study participants were recruited and surveyed between November 2020 and January 2021. Study participants were recruited through online advertisement and neighborhood listservs, with permission from moderators. Participants were eligible if they were adults, 18 years or older, and had tested for COVID-19 within 30 days prior to the survey. At the time of recruitment, no at-home antigen tests were readily available. RT-PCR tests were done at a variety of laboratories while rapid antigen tests were available primarily at point-of-care. Survey participants self-reported their test results. Twenty-two study participants were excluded from the final analysis because their test results were inconclusive or not available in time. The survey was confidential and did not include any participation incentive information. Subjects had to complete a separate survey for receiving a gift card for participation, where their names and phone numbers were recorded. The survey was designed such that participants could not take the survey multiple times. The survey instrument is available in the supplemental digital content. ^[13]

Data were collected prior to vaccination being made widely available, and when Alpha variant of the novel coronavirus was predominant in the U.S. In addition to COVID-19 test results, the survey captured participants’ COVID-19 symptoms and exposures within 30 days prior to the survey, as well as general health status and socio-demographic characteristics. At the time of data collection, the national COVID-19 prevalence was 5%. The sample size of 400 was calculated to include at least 50 individuals who tested positive for an infection causing COVID-19. The subject’s age, gender and race were weighted to reflect those in the United States’ population. The accuracy of symptoms in predicting COVID-19 diagnosis was examined using Area under the Receiver Operating Curve (AROC). Since the likelihood ratio associated with each symptom was calculated based on the data obtained from the literature, the reported accuracy rate is a cross-validated AROC on an independent sample of patients.

This study was approved by George Mason University IRB (number 1668273–8).

Results

In total, 483 participants completed the survey. 22 respondents did not know their test results at the time of participation in the study or had an inconclusive test result. Analysis was done on the remaining 461 patients. Table 1 shows the demographic characteristics of the population recruited for the study. These cases were weighted to reflect the distribution of age, gender, and race in the United States.

Table 1:

Characteristics of Study Sample

COVID-19 Test Results	Number of Cases (%)
Negative	330 (68.32%)
Positive	131 (27.12%)
Results Pending	15 (3.11%)
Inconclusive	7 (1.45%)
Age
18–24	84 (17.39%)
25–34	210 (43.48%)
35–44	156 (32.30%)
45–54	20 (4.43%)
55–84	13 (2.69%)
Gender
Female	279 (57.76%)
Male	203 (42.03%)
Ethnicity
Hispanic Latino	60 (12.42%)
Non-Hispanic Latino	401 (83.02%)
Unknown	22 (4.55%)
Race
Other	18 (3.75%)
Asian	25 (5.18%)
Black or African American	60 (12.42%)
White	380 (78.67%)
Essential Workers
No	282 (58.39%)
Yes	201 (41.61%)
Healthcare Workers
No	281 (58.18%)
Yes	202 (41.82%)

Table 2 provides the symptoms of COVID-19, as identified from the literature review. Table 2 also reports the likelihood ratios associated with each symptom, estimated from sensitivity, specificity or the prevalence of the symptom reported in the literature. A likelihood ratio above 1 indicates how many times the symptom increased the odds of COVID-19 diagnosis. A ratio below 1 indicates symptoms useful for ruling out COVID-19 diagnosis; the smaller the value the more useful it is. Likelihood ratios near 1 indicate symptoms that are neither useful as predictors nor rule out COVID-19 diagnosis. It is noteworthy to report that cough, by itself, was not predictive of COVID-19 diagnosis (LR=0.93); fever was predictive (LR=1.62). Cough and fever were two of the early symptoms used to triage patients at the onset of the pandemic. Later studies also identified loss of smell as a symptom of COVID-19, and its likelihood ratio, LR= 5.5, suggests that it is a strong predictor of COVID-19 diagnosis.

Table 2:

Likelihood Ratios (LR) Associated with Symptoms of COVID-19

Non-Respiratory Symptoms	Number of Patients	Sensitivity	Specificity	LR+	LR−
General
Fever or feeling feverish 14–23	5,484	0.44	0.73	1.62	0.77
Muscle aches/myalgia 12,20,22	1,427	0.30	0.83	1.74	0.85
Pinkeye/Conjunctivitis 1	30,494	0.01	1.00	3.46	0.99
Fatigue (more than normal) 12,14,16,17	273	0.41	0.70	1.37	0.84
Chills 12,16,20	1,443	0.09	0.89	0.84	1.02
Neurological
Headaches 12,14,16,17,20,22	1,700	0.16	0.87	1.24	0.96
Loss of balance17	88	0.38	0.78	1.73	0.79
New confusion 25,26,27, *	3,848			0.42	1.06
Unusual shivering or shaking 12,16	132	0.14	0.86	1.00	1.00
Loss of smell 12,28	262	0.22	0.96	5.50	0.81
Loss of taste 12,27	262	0.20	0.95	4.00	0.84
Seizures 25,29,30,31, *	518			0.12	1.04
Gastrointestinal
Diarrhea12, 14–17,20,22	1,733	0.05	0.95	0.88	1.01
Stomach/abdominal pain 12,14,16	185	0.01	0.97	0.50	1.01
Change in or loss of appetite 32	2,763	0.47	0.64	1.31	0.83
Nausea or vomiting12,14,16,20	489	0.03	0.97	1.13	1.00
Inflammatory
Joint/other unexplained pain (myalgia/arthralgia)12,14–18	339	0.57	0.67	1.74	0.64
Red/purple rash or lesions on toes 33	318	0.87	0.84	5.58	0.15
Unexplained rashes 24	30,494	0.02	0.97	0.68	1.01
Excessive sweating 24	30,494	0.06	0.97	1.68	0.98
Respiratory Symptoms
Cough 12, 14–18,20,34,35	2,607	0.57	0.39	0.93	1.1
Sore Throat 12,14–18,20, 34	5,045	0.09	0.33	0.13	2.73
Difficulty breathing (Dyspnea) 12,14–18,20,33, 34	2,554	0.17	0.84	1.08	0.99
Shortness of breath (Hypoxia) 19	2,929	0.15	0.83	0.88	1.02
Runny nose (Rhinorrhea/nasal symptoms) 12,14–18, 20, 34	5,334	0.04	0.37	0.06	2.57
Chest pain (chest tightness) 12,36	34	0.05	1.00	--	0.95

^*Calculated from prevalence of symptoms.

Combinations of Symptoms

Table 3 reports the accuracy of symptoms and combinations of symptoms in predicting COVID-19 diagnosis. For patients with non-respiratory symptoms, the Area under the Receiver Operating Curve (AROC) ranged from 0.62 to 0.81, a moderate to high level of accuracy. The more non-respiratory symptoms were present the more accurate were the predictions. When two or more sets of non-respiratory symptoms were present, the AROC ranged from 0.72 to 0.81.

Table 3:

Accuracy of Symptom Screening for Patients with Different Clinical Presentations

Non-Respiratory Presentations			& Respiratory Presentation
Category	Cases with at Least 1 Symptom in Listed Categories (Percent)	AROC	Cases with at Least 1 Symptom in Category (Percent)	AROC
None			388 (84%)	0.46
Neurological	266 (58%)	0.64	248 (54%)	0.50
Gastrointestinal	234 (51%)	0.62	221 (48%)	0.47
Inflammatory	134 (29%)	0.67	127 (28%)	0.40
General	378 (82%)	0.73	357 (77%)	0.56
Neurological & Gastrointestinal	189 (41%)	0.74	180 (39%)	0.50
Neurological & Inflammatory	121 (26%)	0.72	116 (25%)	0.48
Neurological & General	241 (52%)	0.73	231 (50%)	0.57
Gastrointestinal & Inflammatory	120 (26%)	0.73	115 (25%)	0.48
Gastrointestinal & General	233 (51%)	0.71	215 (47%)	0.51
Inflammatory & General	127 (28%)	0.72	122 (26%)	0.46
Neurological, Gastrointestinal & Inflammatory	114 (25%)	0.78	110 (24%)	0.52
Neurological, Gastrointestinal & General	180 (39%)	0.77	175 (38%)	0.57
Gastrointestinal, Inflammatory & General	119 (26%)	0.77	114 (25%)	0.50
All	113 (25%)	0.81	109 (24%)	0.60

^*red/purple toes and sore throat were dropped from the analysis as it reduced accuracy of predictions

Patients often present with multiple symptoms. We assume that each symptom has an independent impact on the diagnosis of COVID-19. Under this assumption, the odds of COVID-19 diagnosis are the product of the likelihood ratios associated with each of the symptoms, including symptoms that are absent. If a patient does not have any of the symptoms in the category, then the likelihood ratios associated with absence of the symptoms in that category are ignored.

Table 4 shows an example of a patient, presenting with symptoms in different categories. This Table shows a patient who has no neurological symptoms. When all symptoms in a category are absent, no prediction is made in that symptom category. When one or more symptoms in the category are present, then a prediction is made for the combination of symptoms by multiplying the likelihood ratios associated with the symptom. For the patient in Table 4, the odds of COVID-19 diagnosis are highest in the general category. The odds of COVID-19 diagnosis for this patient were 0.11 (versus prior odds of 0.03), which corresponds to the probability of 0.10 (versus population prevalence of 0.02). The patient is at an elevated risk of having COVID-19. Likelihood ratios modify prior odds of COVID-19 to estimate the patients’ odds of having COVID-19. The estimation of prior odds requires access to local odds of COVID-19. This information is not always available. When unknown, the prior odds of 1 can be assumed, and the posterior odds can be interpreted to the extent of the evidence which supports the presence of COVID-19. ^[37]

Table 4:

An Example Patient with Multiple Symptoms

Non-Respiratory Symptoms	One Patient’s Symptoms	Associated Likelihood Ratio	Impact of Relevant Symptoms	Odds of COVID-19
General
Fever/Feeling Feverish	Present	1.62	4.08	0.11
Muscle Aches/Myalgia	Absent	0.85
Pinkeye/Conjunctivitis	Present	3.46
Fatigue (More Than Normal)	Absent	0.84
Chills	Absent	1.02
Neurological
Headaches	Absent	0.96	None Calculated as All Symptoms Are Absent
Loss of Balance	Absent	0.79
Slurred Speech	Absent	1
New Confusion	Absent	1
Unusual Shivering or Shaking	Absent	1
Loss of Smell	Absent	0.81
Loss of Taste	Absent	0.84
Tingling/Numbness/Swelling in Hands/Feet	Absent	1
Seizures	Absent	1
Gastrointestinal
Diarrhea	Absent	0.88	0.83	0.02
Stomach/Abdominal Pain	Absent	1.01
Change in/Loss of Appetite	Absent	0.83
Nausea/Vomiting	Present	1.13
Inflammatory
Joint/Other Unexplained Pain (Myalgia/Arthralgia)	Absent	0.64	0.10	0.00
Red/Purple Rash/Lesions on Toes	Absent	0.15
Unexplained Rashes	Absent	1.01
Excessive Sweating	Absent	0.98
Prior Odds
Weighted Sample		0.03
Sample		0.40

Discussion

Our analyses showed that cough, sore throat, runny nose, dyspnea, and hypoxia are not good predictors of COVID-19 diagnosis. The Centers for Disease Control and Prevention (CDC) lists cough as a common symptom of COVID-19 that can be used to diagnose COVID-19. One possible explanation for divergence in findings between the CDC and our analyses may be the role that seasonality plays on the diagnostic value of cough. Our scoping literature review identified 29 symptoms that can be used in diagnosing COVID-19, including respiratory, neurological, gastrointestinal, inflammatory, and general symptoms. There was a wide range in values of the likelihood ratios, suggesting that symptoms are not equally important. Some of the symptoms that are commonly used for COVID-19 screening, e.g., cough, showed not to be the best predictors. The literature review showed that individuals with cough were less likely to have COVID-19 than other diseases. This does not imply that cough is uncommon among COVID-19 patients but emphasizes that cough is also common in people with seasonal allergies, influenza and other diseases caused by respiratory pathogens.

The estimated likelihood ratios suggest a group of symptoms that increases the odds of COVID-19 (e.g., loss of smell), another group of symptoms which decreases the odds (e.g., runny nose), and yet another set of symptoms that does neither (e.g., difficulty breathing). This suggests that combinations of different symptoms could play an important role in suggesting and ruling out COVID-19 diagnosis.

Relying solely on the respiratory symptoms (e.g., cough, runny nose, but not fever) to diagnose COVID-19 had low accuracy (AROC of 0.48), which is no better than random guessing. Relying on respiratory and general symptoms (e.g., fever) improved the accuracy to 0.56, but this is too low to be clinically significant. Our analyses demonstrate that patients who present with non-respiratory symptoms can be more accurately evaluated (average AROC of 0.73). For those patients, clinicians could base their triage decisions on the reported symptoms. Even when diagnosing patients with combined non-respiratory and respiratory symptoms, the model accuracy was moderate (AROC=0.60). In a subset of 25% of patients who presented with symptoms in multiple non-respiratory categories and reported no respiratory symptoms, our predictions were highly accurate (AROC of 0.81). Therefore, clinicians can be more confident in their triage of COVID-19 patients that present with non-respiratory or multi-systemic symptoms which do not include respiratory symptoms. If patients present with only respiratory symptoms, other information is needed. For those patients, triage could be delayed until diagnostic testing is performed to either confirm or rule out the infection or COVID-19 diagnosis.

In the community, there are two ways to assess the presence of COVID-19 in a patient. One is based on clinical presentation of the disease and the other is based on diagnostic laboratory testing (i.e., RT-PCR) or at-home testing (i.e., antigen test). Of course, one can do both and many approved antigen tests for COVID-19 require the test results to be assessed in the context of patient’s symptoms and the clinical presentation of the disease. Few studies have compared the performance of symptom screening and antigen at-home testing, as part of the same study. In our additional work, we presented a study that compares the accuracy of antigen tests with symptom screening. ^[38]

Limitations

This study had a number of limitations. First, in examining the impact of the combination of symptoms, we assumed that symptoms were independent. This assumption is obviously false as certain symptoms may co-occur. It may have been more reasonable to diagnose COVID-19 based on the clusters of symptoms, rather than individual symptoms as some of our additional work suggested. ^[39] This would have allowed us to examine dependencies among the symptoms. Unfortunately, due to the examined symptoms having been obtained from different studies as part of the literature review, we could not estimate the likelihood ratios associated with symptoms across different studies.

Second, our study did not look at how seasons affect the diagnostic value of symptoms. The diagnostic value of symptoms may vary during and outside of flu season, further complicating the accuracy of COVID-19 diagnosis.^[40] Flu and allergies are seasonal diseases ^[41] and, depending on the prevalence of those diseases in a given year, cough may play a different role in the diagnosis of COVID-19.

Third, other factors such as age, exposure history, virus variants as well as vaccination status can also affect clinical presentation of COVID-19, as reported in the literature. ^[42,43] Most of the literature cited in Table 2 reported on data collected in the first 9 months of 2020 whereas our survey data were collected in the last 3 months of 2020, when a new viral strain, Delta variant, was beginning to spread. While no distinction in symptoms were known between different viral strains in the first year of the pandemic, we have learned, overtime, that Omicron had symptoms affecting upper airways more than the earlier strains of the virus. ^[44] Hence, our findings show a way to establish a screening tool but the tool itself would need to be calibrated based on new data on prevalence of symptoms and the resulting likelihood ratios.

Conclusion

Nearly 3 years into the pandemic no strides have been made on how to differentiate COVID-19 from other diseases with overlapping symptoms. Unlike other studies that used symptoms for diagnostic and prognostic models of COVID-19 ^[45,46], this paper provides analysis on how to differentiate COVID-19 from other diseases based on the symptoms and the likelihood ratios derived from the prevalence of those symptoms in patients with and without COVID-19. This study also examines the combinations of symptoms that are most predictive of COVID-19 diagnosis. Our data suggests that the types of symptoms that are present, both non-respiratory and respiratory, and particular combinations of those symptoms, are important to consider during screening for and diagnosing COVID-19. Patients differ considerably in their clinical presentation. However, when presenting with any symptoms, especially multi-systemic symptoms, rather than only respiratory ones, the accuracy of predictions improves. Our findings suggest that a symptom screening tool which accounts for diagnostic values of symptoms and various combinations of symptoms could have important clinical implications and utility.