Validating a pre-flight algorithm for predicting in-flight oxygen in patients with interstitial lung disease

Extract

The hypoxic challenge test (HCT) is the most widely used test in clinical practice to determine which patients require in-flight oxygen. The use of pre-flight algorithms is a cost-effective method to help identify which interstitial lung disease (ILD) patients are fit to fly without supplemental oxygen and therefore do not need a HCT. Earlier studies have developed equations to predict arterial oxygen tension (P_aO2) at altitude, this was predominately in a chronic obstructive disease population [1–6].

Shareable abstract

Resting P_aO2 >9.30 kPa is the best predictor of a pass on a hypoxic challenge test (HCT), suggesting that decisions about the need for a HCT, and therefore the ability to fly without supplemental oxygen, can be made from a blood gas reading in ILD https://bit.ly/40Db0bH

To the Editor:

The hypoxic challenge test (HCT) is the most widely used test in clinical practice to determine which patients require in-flight oxygen. The use of pre-flight algorithms is a cost-effective method to help identify which interstitial lung disease (ILD) patients are fit to fly without supplemental oxygen and therefore do not need a HCT. Earlier studies have developed equations to predict arterial oxygen tension (P_aO2) at altitude, this was predominately in a chronic obstructive disease population [1–6]. There have been limited studies investigating which ILD patients are appropriate to refer for a HCT. The most established findings came from Barratt et al. [7] in 2018, which informed the British Thoracic Society Clinical Statement on Air Travel for Passengers with Respiratory Disease (2022) [8]. They developed a novel algorithm which recommends that when both P_aO2 is greater than 9.42 kPa and transfer factor of the lung for carbon monoxide (T_LCO) is >50% predicted, no in-flight supplemental oxygen is required. Alternatively, diagnostic HCT is advised in patients in whom either T_LCO ≤50% or P_aO2 ≤9.42 kPa and when both are reduced, supplemental oxygen is recommended. They acknowledged that validation of their results was necessary.

The primary objective of this study was to determine how accurately the algorithm developed by Barratt et al. [7] predicted a HCT pass in our ILD cohort. The secondary objective was to find better predictors for HCT outcomes using data from our ILD cohort.

This study was approved by the Papworth Cardiorespiratory Physiology Research Database data access committee. Data was compiled retrospectively from 130 ILD patients at Royal Papworth Hospital between 2017–2024. The HCT was carried out using the venturi mask method. Capillary blood gases were used to measure P_aO2 via the earlobe. Pulmonary function tests were conducted as per American Thoracic Society/European Respiratory Society guidelines [9–10] with parameters collected within ±3 months of HCT performance. The 6-min walk test was performed in accordance with American Thoracic Society guidelines [11]. Statistical analysis was performed on Jamovi version 2.42 and STATA version 18. The cut-off thresholds created by Barratt et al. [7], were evaluated by calculating the sensitivity and specificity for passing a HCT. Receiver operating characteristics analysis identified optimal predictors and cut-off thresholds for a HCT “pass” outcome. Binomial multivariate logistic regression identified independent predictors for HCT pass outcome from the data, controlling for demographic factors.

Our ILD cohort was mostly male (70%) with a mean age of 68.1±10.3 years. Gender age physiology (GAP) score: 4 (interquartile range (IQR) 2), indicative of moderate severity. Median T_LCO: 43.5% (IQR 20.5) predicted. Primary diagnosis was idiopathic pulmonary fibrosis (60%). Our ILD cohort had a HCT pass outcome in 90 out of 130 patients (69%) compared to 49% for Barratt et al. [7]. Both cohorts shared similar characteristics, with the majority being male and exhibiting comparable ages, GAP severity, forced vital capacity (FVC) % predicted, T_LCO % pred, and a prevalent idiopathic pulmonary fibrosis diagnosis.

Validating the combined cut-off of P_aO2 >9.42 and T_LCO >50% developed by Barratt et al. [7], the sensitivity was 31.5% and specificity was 96.7% for predicting a HCT pass. This combined cut-off value accurately predicted 33 of 130 patients to pass the HCT. When validating P_aO2 (>9.42) and T_LCO (>50%) cut-off thresholds separately, sensitivity was 77.5% and 38.4% respectively while specificity was 38.4% and 93.5% respectively.

In our cohort, we found that that resting P_aO2, T_LCO % and FVC % had the best area under the curve (AUC) of 0.84, 0.75 and 0.72 respectively, for predicting a HCT pass outcome. The optimal cut-off values for P_aO2 kPa and T_LCO % and FVC % determined from the receiver operating characteristics analysis were included in the regression model. A successful HCT outcome was independently predicted by resting P_aO2>9.30 kPa (OR 11.97, 95% CI 2.46–58.2), T_LCO >43% predicted (OR 36.4, 95% CI 5.58–236.8) FVC >60% (OR 5.55, 95% CI 1.20–25.8) and body mass index (OR 0.77, 95% CI 0.72–0.95) The model demonstrated an AUC of 0.94, specificity of 79% and sensitivity of 92%.

Comparing our developed P_aO2 >9.30 threshold with Barratt's P_aO2 >9.42 threshold, both showed comparable sensitivity of 81.1% and 77.8% respectively, and specificity of 72.5% and 77.5% in predicting a HCT pass, as well as similar AUC (0.78 versus 0.76 respectively p=0.675). The P_aO2 >9.42 threshold correctly predicted 101 patients to pass HCT, in comparison to the 105 patients predicted by using the P_aO2 >9.30 threshold. Overall, the P_aO2 >9.30 threshold had the highest sensitivity (81.1%) for predicting a HCT pass.

Comparing our developed T_LCO threshold of >43% compared with the validated T_LCO >50% threshold, the T_LCO >43% threshold exhibited a higher sensitivity of 64.4% compared to 38.4% for T_LCO >50%, and a marginally worse specificity of 87.1% and 94.6% respectively. The AUC for T_LCO >43% (0.76) was significantly higher (p=0.004) than T_LCO >50% (AUC 0.66). In predicting HCT pass, T_LCO >43% correctly identified 67 of 107 patients, more than T_LCO>50% which only correctly identified 40 of 107 patients.

We devised a much-simplified pre-flight algorithm based on the P_aO2 >9.30 threshold as this demonstrated best ability to predict a HCT pass (see figure 1).

FIGURE 1.

Interstitial lung disease (ILD) pre-flight algorithm. ILD patients with arterial oxygen tension (P_aO2) >9.30 kPa can travel without supplemental in-flight oxygen. Patients with P_aO2 ≤9.30 kPa require a hypoxic challenge test (HCT) to determine if in-flight oxygen is required.

Our study validated the pre-flight algorithm from Barratt et al. [7] (P_aO2>9.42 and T_LCO>50%), which showed high specificity and low sensitivity for predicting a HCT pass. Our findings suggest the combined algorithm sensitivity for predicting a pass was tied to the T_LCO (>50%) threshold. As a notable proportion of patients achieved a HCT pass outcome despite T_LCO<50%, if the intention of the algorithm is to identify patients who pass a HCT, then our findings suggest adding T_LCO% creates unnecessary ambiguity at the cost of sensitivity.

Our results indicated that P_aO2 >9.30 demonstrated the best sensitivity in predicting a HCT pass. This cut-off threshold comparable to the P_aO2 (>9.42) threshold developed by Barratt et al. [7], which predicted a HCT pass with a similar level of sensitivity. Furthermore, the Aerospace Medical Association advises that flying without oxygen supplementation is considered safe if the pre-flight P_aO2 is greater than 9.33 kPa [12]. Based on the P_aO2 >9.30 criteria, we created a pre-flight algorithm that has an 81% sensitivity for predicting HCT pass. The advantage of our developed pre-flight algorithm is that is a cost-effective method to help identify which ILD patients are fit to fly without supplemental oxygen and therefore do not need a HCT. The other advantage of our algorithm over the extant algorithm developed by Barratt and colleagues is that T_LCO is not required, which is not available in all settings.

This study had number of strengths. We are the first study to validate the pre-flight algorithm developed by Barratt et al. [7] in another ILD cohort. Furthermore, our multivariate analyses accounted for demographic factors unlike those done by Barratt et al. [7]. The retrospective study design limited our ability to link results with actual flight outcomes reported by patients. As with all retrospective analyses, there is possible bias from missing data as not all patients had complete lung function and 6-min walk test data. Furthermore, analyses were not controlled for influence of comorbidities such pulmonary hypertension and COPD due to insufficient cases. Finally, HCT is not the gold standard test for determining in-flight oxygen compared with hypobaric chambers, which are not widely available. Additionally, the short duration of a HCT may not be sufficient time to detect in-flight hypoxia.

To conclude, Resting P_aO2 of >9.30 kPa alone best predicts HCT pass. This needs to be independently prospectively validated against flight outcomes.