A total of 134 patients were invited for follow-up (10 patients declined follow-up, 13 patients were lost to follow-up and 10 patients did not attend for chest radiographs). Therefore 101 patients who had complete data including paired baseline and follow-up chest radiographs were included in the final analysis (study CONSORT diagram see Supplemental Figure 1).
Baseline demographics for the cohort of 101 patients (Table 1) identified 53.5% were males with a median (IQR) age of 53.0 (45-63) years. The median baseline to follow-up chest radiograph interval, and admission to virtual follow-up appointment, was 82 (77-86) days and 96 (94-98) days respectively. The median length of stay was 9 (5-17.5) days and 48.5% (n=49) of patients were admitted to a level 2 or 3 care facility. The majority of patients were white British ethnicity (65%) and never smokers (65%). Common comorbities included hypertension (36%), obesity [BMI>30 kg/m2] (28%) and diabetes melitus [all types] (15%).
At 12-week follow-up 65% (n=65) of patients had one or more persisting symptom of which fatigue (41%) and breathlessness (38%) were the most common. 55% (n=55) of patients were discharged following their 12-week virtual clinic. In patients who required additional follow-up (n=40), planned further investigation included; pulmonary function testing (88%), transthoracic echocardiogram (78%), chest computed tomography (CT) scanning (33%), repeat blood tests 20%, and repeat chest radiograph (8%).
Chest Radiograph Scoring
The median (IQR) baseline chest radiograph score was 4.0 (3-5) with a maximum score of 8. At follow-up the median (IQR) chest radiograph severity score was 0 (0-1) with a maximum of 7. Overall 31.6% (n=32) of patients had persistent chest radiograph changes at follow-up. Of those with persistent chest radiograph change, 55% (n=17) had an infiltrate score of 1, 22% (n=7) had a score of 2, and 23% (n=8) had a score of 3 or greater. There was no significant difference in the median chest radiograph follow-up interval between those with resolution and persistent radiograph abnormality; 83.0 days (76.0-99.0) versus 81.5 days (65.5-87.5) respectively, p=0.37. To date 11 patients in the cohort have proceeded to have chest CT scans, 8 of which have identified evidence of pulmonary fibrosis (for description see Supplemental Information).
Predicting persistent chest radiograph abnormality
Patients with abnormal chest radiographs at follow-up had a significantly longer median length of stay (20.5 days vs. 8.0 days p<0.01) and were more likely to be current or previous smokers compared to never smokers (56% vs. 23% p=0.02). Furthermore, patients with persistent radiograph abnormality were more likely to have been admitted to a level 2 or 3 care facility compared to those with complete radiograph resolution (45% vs. 19%) p<0.01. Correlation was observed between total inpatient and follow-up chest radiograph scores r=0.340 p=0.001. There was no significant difference between groups in the distribution in the use of oxygen, by past medical history, ethnicity or in the follow-up interval between inpatient and follow-up chest radiographs (Table 1).
Univariate regression analysis identified length of stay, age, and current or ex-smoker status as significant risk factors for persistent chest radiograph change (Table 2a). In multivariate analysis (covariates; length of stay, age at admission, level 2 or 3 admission, current or ex-smoking status and obesity) length of stay, current or ex-smoking status and obesity were identified as independent risk factors for persistent chest radiograph abnormality Table 2b.
Analysis of Laboratory Indices
Baseline and follow-up pathology data are summarised in Table 3a and 3b. This analysis identified that serum lactate dehydrogenase (LDH) was the only blood marker signficantly elevated in patients with persistent chest radiograph abnormality compared to those with complete resolution at both baseline and follow-up; median (IQR) 959 U/L (697-1193) vs. 679 U/L (502-955) p=0.006 (n=76) and 435 U/L (359-489) vs. 373 U/L (317-405) p=0.004 (n=46) respectively Figure 1a and 1b. Further analysis of LDH results identified significant correlation between baseline serum LDH levels and both baseline and follow-up chest radiograph scores (r=0.34, p=0.003 and r=0.34 p=0.003 respectively). Follow-up serum LDH levels correlated significantly with follow-up chest radiograph scores (r=0.39 p=0.007). The baseline serum LDH was also signficantly correlated with length of stay (r=0.43 p<0.001).
In univariate analysis, baseline serum LDH demonstrated a strong trend of association with increased risk of persistent chest radiograph abnormality but failed to reach statistical significance Hazard Ratio (HR) 1.001, 95% Confiedence Interval [95%CI] (1.000-1.005) p=0.078). However, in a multivariate model correcting for obesity and current or ex-smoking status, baseline serum LDH was a significant predictor for persistent chest radiograph abnormality, HR 1.001 (95%CI 1.000-1.002) p=0.046.
Area Under the Receiver Operating Characteristic Curve (AUROC) Analysis
Using the above identified risk factors we sought to identify a combined variable tool that predicted patients most at risk of developing persistent chest radiograph abnormality at 3-months. Using area under the receiver operating curve (AUROC) analysis, a combined variable analysis was conducted. A combined score with one point each for; length of stay ≥ 15 days, LDH ≥ 750 U/L, positive current or ex-smoking status, admission to a level 2 or 3 care facility, and positive past medical history of obesity (BMI> 30kg/m2) was identified as providing the best prediction model for risk of persistent chest radiograph abnormality with an AUROC of 0.808 (95%CI 0.701-0.915) p<0.0001 Table 4a and Table 4b, and Figure 2. With a cut off of 3 points or more demonstrating an AUROC of 0.76 (95%CI 0.625-0.885) with of 0.70 sensitivity and a specificity of 0.81.
Applying this same risk prediction score to patients outcome destination from the 12-week virtual clinic i.e discharge (n=55) or need for a further appointment (n=46) identified that this risk calculator still performed strongly, predicting patients who needed further follow-up appointments with an AUROC of 0.781 (95%CI 0.664-0.898) p<0.0001) Table 4c.