COVID-19 TARRACO Cohort Study: Development of a predictive prognostic rule for early assessment of COVID-19 patients in primary care settings

INTRODUCTION

The novel coronavirus SARS-COV-2 infection is causing the greatest infectious pandemic disease (COVID-19) in this century, with more than 60 million infected persons to date.(1-3) Despite some time elapsed since pandemic started, population-based data about susceptibility and risk factors for COVID-19 is limited. There are numerous studies reporting clinical features and outcomes among hospitalised patients with severe COVID-19, but community- or population-based data including the overall spectrum of disease (i.e., patients with mild or moderate symptoms managed as outpatient or in ambulatory basis) is scarce.(1,3)

It seems well established that the most severe and fatal cases, although they can occur in young and “healthy” people, are more frequent in older people and / or with comorbidities, like cardiovascular disease, hypertension, diabetes, chronic respiratory disease or chronic kidney disease.(4-6) Smoking, obesity, male sex, socioeconomic deprivation and ethnicity were also associated with greater severity/lethality.(7,8) Lastly, multiple laboratory findings, viral and genetics factors may be associated with worse outcomes.(9,10)

Several prognostic rules/algorithms have been reported by combining demographic, clinical, radiological and laboratory data.(11-16) However, since a family physician point of view, the majority of predictive markers are not very useful in primary care settings because most of them requires radiological and/or analytical data few available in ambulatory basis.(17)

This study investigated the possible relationship between pre-existing conditions/comorbidities and onset symptoms with the risk of suffering poor outcomes in a population-based cohort of community-dwelling middle-aged and older adults with symptomatic COVID-19 (including hospitalised and ambulatory cases) in the region of Tarragona (old called Tarraco, Spain). In addition, we explored and developed simple COVID-19 screening rules to predict critical outcomes (ICU admission or death) useful in outpatient settings.

METHODS

RESULTS

Of the 345 overall laboratory-confirmed COVID-19 cases observed in non-institutionalised cohort members across the study period, 63 were excluded from this analysis because non-available clinical data (n=2), to be asymptomatic cases (n=27) or to be considered nosocomial cases occurred inside social-health hospital stay (n=34). Thus, the remaining 282 community-dwelling/non-institutionalised symptomatic COVID-19 cases were finally included in this report.

Mean age of cases was 65.9 years (SD: 12.1) and 140 (49.6%) were men. One hundred and fifty-four cases were hospitalised (30 of them were admitted in the ICU) and 128 were managed as outpatient (38 of them were attended in the emergency room but were not hospitalised). A total of 45 patients died (two in the emergency room before hospitalisation, 32 in hospital floor and 11 in the ICU). Overall, 64 (22.7%) suffered a critical outcome (ICU-admission or death).

The most prevalent comorbidities were hypertension (42.6%), obesity (25.5%), cardiac disease (23.4%), diabetes (19.1%) and respiratory disease (17.4%).

The most common signs/symptoms at COVID-19 presentation (first 5 days after onset of symptoms or medical contact) were fever (70.9%) and cough (67.7%) followed by general discomfort (42.2%), dyspnea (39.7%), fatigue (30.1%), myalgias (27.3%), diarrhea (24.1%) and headache (23%).

Table 1 compares demographic and clinical characteristics of cases according if they presented or not a critical outcome. Patients who suffered a critical outcome were older (mean age 73.6 vs 63.7 years, p<0.001) and had more comorbidities (2.6 vs 1.5 comorbidities, p<0.001) than those who did not suffer a critical outcome.

In the crude analyses several pre-existing conditions/comorbidities (renal, respiratory, cardiac disease, diabetes and hypertension) and two symptoms (dyspnea and confusion) were significantly associated with a greater risk of admission to the ICU/death, whereas other symptoms (myalgias, headache and anosmia) were associated with a reduced risk. After age and sex-adjustment, apart from increasing age/years (OR: 1.07) and amle sex (OR: 1.64), only chronic respiratory disease (OR: 2.20) and diabetes (OR: 1.89), dyspnea (OR: 5.30) and confusion (OR: 4.95) remained significantly (or nearly significantly) associated with an increased risk, whereas myalgias (OR: 0.27) and the cluster symptom combining ageusia/anosmia/rinorrhea (or: 0.36) were associated with a significant reduced risk (table 2).

Finally, after multivariable-adjustment only age/years (OR: 1.04; 95% CI: 1.01-1.07; p=0.004), confusion (OR: 5.33; 95% CI: 1.54-18.48; p=0.008), dyspnea (OR: 5.41; 95% CI: 2.74-10.69; p<0.001) and myalgias (OR: 0.30; 95% CI: 0.10-0.93; p=0.038) remained significantly associated with increased or reduced risk of suffering a critical outcome (Table 3).

We used this multivariable-adjusted model to develop a simple predictive prognostic rule, named CD65-M (acronym for Confusion, Dyspnea, and Age >65 years, which contributed with a positive point each; and Myalgias which contributed with a negative point in the score), that showed an acceptable correlation with the risk of suffering a critical outcome (area under ROC curve: 0.828; 95% CI: 0.774-0.882).

Additionally, by using the logistic regression model adjusted by age and sex (table 2) we explored and developed a larger predictive rule including the 8 significant variables in that model. According to magnitude changes in the ORs, this rule called CD65RD-WMA (acronym of the initials of the names of the 8 included variables) weights 2 points for each one of Confusion and Dyspnea, 2 points for age greater than 80 years and one point for age between 65-79 years, one point for Respiratory disease and other for Diabetes. Furthermore, the score weights a negative point for each one of the following: Women, Myalgias and combined Ageusia/anosmia/rinorrhea. This longest CD65RD-WMA score (which may vary between −3 and 8 points) also showed a good correlation with the probability of suffering a critical outcome during clinical course, but did not substantially increased discriminatory power (area under ROC curve: 0,836; 95% CI: 0.784-0.889)

Figure 1 compares ROC curves for both CD65 and CD65RD-WMA rules.

• Download figure
• Open in new tab

Figure 1.

Comparison of ROC curves for both simple CD65-M and longer CD65RD-WMA rules.

Table 4 shows the absolute numbers and percentages of observed cases and critical outcomes in each scoring for both simple and longer rules.

Sensitivity, specificity, positive and negative predictive values for different cut off points in both rules are shown in Table 5.

DISCUSSION

In the current context of global pandemic and public health emergency by COVID-19, the knowledge of protective and/or risk factors to suffer infection or develop poor outcomes is essential for adequate management of patients in earlier stages of the disease.(17) In the present community-based cohort study, apart from increasing age that has repeatedly documented as major risk condition,(4-8) we have also found that some early symptoms (within the first five days after the onset of disease) are associated with better or poor prognosis (i.e, need of ICU admission or death). Indeed, presence of dyspnea and confusion clearly increased the risk of ICU-admission or death, whereas myalgia was associated with a reduced risk.

On the first, it is not surprising since both variables, or related-variables, were already associated with an increased risk of death in patients with community-acquired pneumonia, having been already included confusion and taquipnea as predictor variables within simple prognostic rules (such as CURB-65 or CRB-65 rules) assessing severity in patients with pneumonia.(22-24)

On the second, surprisingly, myalgias emerged independently associated with a reduced risk for critical outcomes after multivariable adjustments. This may be considered as an unexpected result, but we note that it is not unique and has already been reported by other studies. Concretely, in a single-centre French cohort study that included 150 hospitalised COVID-19 consecutive patients between March-April 2020, the percentage of cases who had myalgias was significantly greater among those patients with favourable clinical course than in those who required ICU-admission or died.(16)

Thus, we constructed a simple prognostic rule containing the above mentioned four variables (CD65-M score), which showed a good correlation with the probability of suffering a critical outcome across clinical course, being 2%, 4.9%, 23.7%, 53.7% And 83.3% for scoring −1, 0, 1, 2 and 3, respectively. According this CD65-M rule, the best cut-off point for predicting bad evolution would be scoring ≥1 (92% sensitivity with 59% specificity) or scoring ≥2 (64% sensitivity with 85% specificity).

If we consider pre-existing comorbidities, many studies have reported increased risk of poor outcomes among patients with diverse comorbidities (especially cardiac, respiratory or renal disease, diabetes, obesity and hypertension), but most studies assessed crude associations (without age- and/or multivariable-adjusted risk).(4-8,25-29)

In the present study we also observed these associations in the crude analyses, but after multivariable adjustment none pre-existing comorbidity was associated with an adjusted increased risk in our study population.

Nevertheless, since respiratory disease and diabetes were significantly associated with an increased risk in the age and sex-adjusted model, we explored and developed a second lengthy rule (CD65RD-WMA) including significant variables in that model. However, this second model did not substantially improve predictive performance of the first simpler CD65-M rule (containing exclusively the 4 significant predictor variables after multivariable-adjustment).

Main strengths in this study were its population-based design (including as hospitalised as well as ambulatory COVID-19 cases), the assessment of early COVID-19 symptoms as possible predictors (rare in the literature), and the large follow-up necessary to adequately recruit the occurrence of critical outcomes during the clinical course (i.e, deaths occurred in hospitalised COVID-19 patients several weeks after hospital or ICU admission). In addition, we performed multivariable-adjusted analysis to accurately estimate relationships between baseline covariables and risk of suffering a critical outcome, developing simple predictive prognostic rules (which may be especially useful in primary care settings). The observed prevalence for most symptoms fits with data reported in other epidemiological studies,(30) which supports the validity of collected data despite retrospective design.

As major limitations, our study was conducted in a single geographical area with specific epidemic conditions (relatively low incidence of COVID-19) in a limited time period (first wave of pandemic) and focused exclusively on community-dwelling people over 50 years.(18) We don’t know the possible influence that changes in study population and/or epidemic intensity could have on the statistics reported here. We also note that the relatively little sample size (282 cases with 64 outcomes in our study) building prediction models may increase the risk of overfitting the model, which implies that the performances of the models in new samples could be worse. We underline that a further validation in an external cohort is necessary before a routine use of the described prognostic rules may be applied. We note, however, that prediction models are needed to support medical decision managing COVID-19 patients.

There are several published or preprint reports that developed prognostic rules predicting critical outcomes (need of mechanical ventilation and/or death), but all of them were based on complementary explorations or laboratory findings which are not generally available in ambulatory or primary care settings.(17) Main contribution of this study lies in the fact that it reports simple clinical prognostic rules including easily accessible clinical data (such as demographics, pre-existing comorbidities and early symptomatology), which could be very helpful assessing COVID-19 patients outside the hospital.

We conclude that the clinical course of COVID-19 patients remains unpredictable in earlier stages of the disease, but simple clinical tools as the proposed CD65-M prognostic rule (pending external validation) may be helpful assessing these patients in primary care settings. Further evaluations adding other easily accessible complementary data (e.g,, pulse-oximetry to detect silent hypoxia) could improve predictive performance.

CONTRIBUTORS

AVC designed the study; AVC and ESG assessed outcomes and wrote the manuscript; ESG, CDC, MFP and IHG obtained data; ESG and AVR did statistical analyses and edited the manuscript; OOG revised the final version; AVC coordinated the study.

FUNDING

This study is supported by a grant from the Instituto de Salud Carlos III of the Spanish Health Ministry (file COV20/00852; call for the SARS-COV-2/COVID-19 disease, RDL 8/2020, March 17, 2020). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript

COMPETING INTEREST

All authors, none declared.