This observational study was conducted in a French University Hospital (North Hospital, APHM, France). The study conformed to the general data protection regulation chart and was registered on the French health data registration portal under #PADS20-207. Patients with persistent clinical impairments after the acute infectious disease stage were selected from a medical consultation designed to assess the quality of recovery after COVID-19. To be included in our study, patients had to have suffered from severe COVID-19 pneumonia (the patients had been admitted to the hospital to receive supplemental oxygen or were submitted to mechanical ventilation). The diagnosis of COVID-19 pneumonia had to be supported by a clinical picture including respiratory difficulties associated with a radiologic pattern-compatible image of pneumonia and SARS-CoV-2 infection confirmed by PCR test. The medical consultation, pulmonary function testing (PFT), and diaphragm ultrasound were scheduled between 3 and 4 months after the hospital discharge for the acute phase of COVID-19 and were undertaken in the PFT lab of the North Hospital.

The medical consultation was designed to identify patients at risk of DD in the screened population. To detect patients at risk for DD, specific parameters were recorded from the patient’s medical history such as a history of trauma, surgery, and neurological, cardiac, or respiratory diseases. The clinical impairments experienced by the patients during the COVID-19 acute phase were also recorded. To assess the severity of the COVID-19, the need for ICU admission, mechanical ventilatory support, and the duration of the support were examined. A chest CT scan was performed at a date close to the medical consultation. The lesions reported by the CT scan and the radiological criteria of severity, graded according to the French Radiology Society guidelines,¹ were recorded [absent, 0; minimal 1- (<10%); moderate, 2- (10–25%); extensive, 3- (25–50%); severe, 4- (50–75%); and critical, 5- (>75%)].

Lastly, the questionnaire and the clinical examination investigated the persistent clinical impairments such as dyspnea, cough, pain, decreased physical capacity, disorders suggesting a neurological condition, and any other medical issues. The dyspnea severity was assessed with the modified Medical Research Council (mMRC) and Borg10 scales.

The pulmonary function test (PFTs) included spirometry and body plethysmograph to measure the vital capacity (VC) and the total lung capacity (TLC) (PFT MasterLab Jaeger plethysmograph, Bunnik, Netherlands). The lung gas diffusion capacity of carbon monoxide (DLCO) was measured using the single-breath method with helium dilution (10). Absolute values were compared to the lower limit of normal (LLN), and the mean values were predicted by the CECA 93 equations (11). The lung functional impairment was screened according to the ATS/ERS definitions (12).

The ultrasonographic examinations were carried out by two experienced investigators (AB and JF), both of whom had performed more than 500 ultrasound examinations of the diaphragm before the beginning of the study. The investigator performing the ultrasound was blinded to the results of the medical consultation and the PFT results. Diaphragmatic function was assessed as both the motion and the thickness of the two hemidiaphragms. The ultrasound examinations were performed using a commercially available ultrasound machine (Vivid S60N, GE Medical System, Milwaukee, WI, United States) equipped with a cardiac probe (3Sc probe) for the diaphragm excursion measurements and a linear vascular transducer (9L probe) for the diaphragm thickness measurements. The examinations were performed with the patients in a seated position.

The excursions of both hemidiaphragms were measured using M-mode, as previously reported (13). Briefly, the probe was positioned on the subcostal or low intercostal area between anterior and posterior interaxillary lines to visualize the right and left hemidiaphragms. The selection of the best incidence was first determined using two-dimensional mode (B-mode). The line was positioned to reach the posterior part of each hemi-diaphragm before applying the M-mode. For a perpendicular approach, anatomical M-mode was used.

The diaphragmatic motion was assessed under three conditions: during quiet breathing at tidal volume, during voluntary sniffing, and during a deep inspiration at total lung capacity. After proper placement of the calipers, the inspiratory diaphragm excursions were measured. Measurements were averaged from at least three different respiratory cycles, except for deep breathing, for which we selected the maximum excursion among several recorded maneuvers.

The right and left hemidiaphragms were visualized below the phrenico-costal sinus near the anterior or the mid-axillary line at the eighth or ninth intercostal space, where the diaphragm abuts the rib cage (zone of apposition) (14, 15). The thicknesses of both hemidiaphragms were measured directly from the frozen B-mode images as the distance from the pleural membrane to the peritoneal membrane, at the end of expiration and at the end of a deep inspiration.

The thickening fraction (TF) was calculated as the following ratio: the thickness at the end of deep inspiration – the thickness at the end of expiration divided by the thickness at the end of expiration.

Diaphragm dysfunction (DD) was diagnosed based on previously published ultrasound criteria (16).

In patients suffering from DD, a follow-up including a medical consultation, PFTs, and ultrasound examination was scheduled 6 months after the first assessment.

The characteristics of the patients suffering from DD (anthropometric data, severity of the acute episode of COVID-19, prior diseases and comorbidities, residual clinical impairments at the time of the medical consultation, and PFT results) were compared with the population with normal diaphragm function.

Numerical data were compared with Students’ t-test. When the data were not normally distributed, a Mann-Whitney test was used.

For qualitative data such as comorbidities and past medical histories, the comparison between groups (i.e., between patients with normal diaphragm function versus patients with DD) was performed using a chi-squared test. Yate’s correction was used when small numbers were involved.

We then searched for the factors associated with DD, using a logistic regression model. To select variables for the final multivariate model, the Boruta random forest method (21) was used for all the variables in the dataset. Variables labeled as confirmed by the Boruta algorithm were retained. Previously, on this dataset, missing data imputation was performed with the missForest non-parametric method (22). Variables with more than 25% missing data were removed.

After Boruta selection, a simple algorithm that computes the Variable Inflation Factor (VIF) was performed on the selected variables. When the maximum VIF value among the variables was greater than two, the corresponding variable was excluded. The process was then repeated until the VIF score of every variable was less than two. This step is crucial to limit the collinearity between the explanatory variables. A logistic model was computed using the selected variables (after Boruta and VIF selection), with the variable to be explained as “abnormal diaphragm function (yes/no)”.

Differences between groups were considered significant at p < 0.05.

Of the 296 patients initially screened, 132 (85 men and 47 women) met the selection criteria and were included in the study (Figure 1). Their mean age was 56 ± 11 years, their mean height was 169 ± 9 cm, their mean weight was 80 ± 17 kg and their mean body mass index was 27.8 ± 5 kg/m².

Forty-three (32%) were obese patients. The mean estimated weight loss induced by the SARS-CoV-2 infection as assessed at the study consultation was 4 kg.

The severity of the clinical status led to ICU admission for 58 patients (44%). Thirty-three patients required mechanical ventilation (25%), with a mean duration of 22 days.

Prior to their SARS-CoV-2 infection, 44 of the patients (33%) suffered from hypertension, 24 (18%) had other cardio-vascular diseases, 35 (26%) had diabetes, 18 (14%) had sleep apnea, and 11 (8%) had COPD.

Additionally, a history of chest trauma was found in three patients (2%), thoracic or cervical surgery in seven cases (5%), and abdominal surgery in 10 cases (8%).

Persistent respiratory difficulties were recorded in 103 patients (77%). The patients reported dyspnea in 87 cases (66%), cough in 32 cases (24%), and chest pain in 31 cases (23%). The other impairments were a decrease in physical capacity secondary to muscular weakness or pain in 34 cases (26%), dysesthesia in 42 cases (32%), cognitive disorders in 23 cases (17%), and palpitations in five cases (4%).

Useable data were obtained from the PFTs in 92% of cases (122 out of 132). The results corresponded to normal function in 53 patients (43.5%), while a restrictive pattern was found in 35 patients (29%). The combination of low diffusion associated with a restrictive pattern was found in 14 cases (11.5%), an isolated low DLCO value was recorded in 15 cases (12%), and a mild obstructive pattern in five patients (4%).

Table 1 lists the results of the population studied.

The ultrasound examination detected 13 cases of abnormal diaphragm function, corresponding to hemidiaphragm paralysis in two patients and DD in 11 others, including six patients with mild dysfunction (four on one side and two on both sides) and five patients with severe dysfunction (3 cases on one side and 2 cases on both sides).

Figures 2, 3 illustrate the recording of diaphragmatic motion in a man suffering from left hemidiaphragm dysfunction.

The chest ultrasonography, the PFT results, and the medical history of the 13 patients with abnormal diaphragm function are presented in Table 2. In five of these 13 cases, risk factors for DD were recorded from the pre-existing medical history.

In one patient, it was possible to ascertain that the hemidiaphragm paralysis appeared after COVID-19, given the normal diaphragm position on the first chest X-ray upon hospital admission followed by abnormal elevation of the dome of the right hemidiaphragm on the post-discharge X-ray checkup. In the other patient, it was not possible to date the paralysis onset, but the CT scan performed at the acute phase of the infection revealed a diaphragm hernia, suggesting it probably arose prior to the COVID-19.

Pleural effusion was recorded in six patients. It was minimal and bilateral in two patients with DD, minor and on the left side in three patients with normal diaphragm function and in one patient with DD.

Men had larger excursions at deep breathing than women, and their hemidiaphragms were thicker on both sides (Table 1).

Tables 3, 4 list the results of the comparison between patients suffering from hemidiaphragm dysfunction or paralysis (DD) versus patients with normal diaphragm function (ND) after COVID-19 severe pneumonia (univariate statistical analysis). The percentage of men was not significantly different between the groups (69% in the DD group vs. 64% in the ND group). The percentage of ICU admissions was not significantly different between the DD (54%) and the ND patients (43%). In the DD group, four patients out of 13 (31%) received mechanical ventilation compared to 29 out of 119 (24%) in the ND group (NS).

A chest CT scan could be performed on a date close to the medical consultation in 91% of the patients (120 out of 132 patients, 108 in the ND group, and 12 in the DD group).

The degree of lung parenchyma impairment as assessed by CT did not differ significantly between the groups (median 1 [1–2.25] in the DD group and 1 [0–2] in the ND group). The frequency of radiologic abnormal imaging was similar concerning ground-glass opacities (62 vs. 52%), consolidation (23 vs. 12%), and reticulation (31 vs. 29%) in the DD and the ND groups, respectively.

Patients with DD more frequently had a history of cardio-thoracic invasive procedures (including cardiac surgery, thoracic surgery, pleural effusion drainage, and atrial fibrillation ablation) or upper abdominal surgery (46% in total) than patients with normal diaphragm function (10%).

The results of the PFTs revealed significantly lower lung volumes and DLCO in patients with DD than in patients with normal diaphragm function. Furthermore, the dyspnea scores (mMRC and Borg10) were significantly increased in patients with DD.

Logistic regression analysis identified mMRC and TLC as the main factors associated with DD (Table 5).

In our population, all patients suffering from DD were managed by respiratory physiotherapy. A checkup visit could be performed for 9 out of these 13 patients. This consultation revealed significant improvement in diaphragm function in 7 out of 9 patients, at 6 months after the first ultrasound examination. This improvement led to normal diaphragm function in four patients, of whom two fully recovered normal PFT values.

As the detection of DD was infrequent in our study, the results of the statistical analysis should be interpreted with a degree of caution. Furthermore, in our population, the pathogenesis of DD most frequently remained hypothetical. Ultrasonography is an appropriate tool for detecting DD in combination with PFT. Nevertheless, to better understand the pathogenesis of the DD, it would be interesting to perform more extensive screening including assessment of phrenic nerve conduction and diaphragm muscle electromyography.