Abstract
Background
Corticosteroids are a mainstay of treatment for acute exacerbations of idiopathic pulmonary fibrosis (AE-IPF), but dosing practices vary. We leveraged between-hospital variation in propensity to administer pulse dose corticosteroids to determine the association between pulse dose corticosteroids and patient outcomes in AE-IPF.
Research Question
Do outcomes differ for hospitalized patients who recieve pulse dose corticosteroids vs low dose corticosteroids for AE-IPF?
Study Design and Methods
We designed an instrumental variable study informed by a target trial framework using the Premier Healthcare Database (2016-2022). We identified adults ≥ 50 years of age with AE-IPF who received a dose of IV methylprednisolone within 2 days of admission. Our exposure of interest was receipt of pulse dose methylprednisolone (≥ 250 mg); our instrument was admission to a hospital with high use of pulse dosing. We assessed association with in-hospital death/discharge to hospice and discharge home without invasive mechanical ventilation. In subgroup analyses, we tested for interaction with unit of admission (admission to ICU/intermediate care units vs ward units).
Results
We identified 3,049 patients with AE-IPF at 177 US hospitals (pulse dose: n = 1,094; low dose: n = 1,955). Patients who received a pulse dose vs low dose had a risk difference for death/hospice of 1.2% (95% CI, −6.0% to 8.5%); for discharge home without invasive mechanical ventilation, this was 5.3% (95% CI, −2.6% to 13.1%). In subgroup analysis, receipt of pulse dosing was associated with differential risk of death/hospice by admission unit (risk difference: ICU/intermediate care: 26.3%; 95% CI, 10.1%-42.6% vs ward units: 0.1%; 95% CI, −11.1% to 11.3%; P interaction = .009), but not with differential risk of discharge home without invasive mechanical ventilation (P interaction = .18).
Interpretation
We observed no significant benefit or harm associated with the receipt of pulse dose corticosteroids for patients with AE-IPF. However, in the subgroup of patients admitted to ICU/intermediate care, there was an increased risk of in-hospital death/discharge to hospice. Future studies should explore the use of pulse vs low-dose corticosteroids in critically ill populations with AE-IPF.
Key Words: idiopathic pulmonary fibrosis, acute exacerbations, corticosteroids, instrumental variable, target trial emulation
Take-Home Points.
Study Question: Does receipt of pulse dose methylprednisolone (≥ 250 mg), compared with lower-dose methylprednisolone, affect outcomes among patients hospitalized with acute exacerbations of idiopathic pulmonary fibrosis (AE-IPF)?
Results: In an instrumental variable analysis of 3,049 patients with AE-IPF across 177 US hospitals, pulse dosing was not associated with overall differences in death or discharge to hospice or discharge home without invasive mechanical ventilation.
Interpretation: Pulse dose corticosteroids were not associated with overall harm or benefit in AE-IPF.
Idiopathic pulmonary fibrosis (IPF)—a chronic, progressive, fibrosing lung disease—is the most common interstitial lung disease, with a poor prognosis.1 Among patients with IPF, acute exacerbations of idiopathic pulmonary fibrosis (AE-IPF) are a leading cause of hospitalization and death.2,3 In-hospital mortality rates for patients with AE-IPF range from 10% in all admitted patients up to 50% in critically ill populations requiring invasive mechanical ventilation (IMV), and the median survival after exacerbation is estimated at 3 to 4 months.2,4,5 Despite high morbidity and mortality, treatment options beyond supportive care are limited.
To date, corticosteroids are the mainstay of treatment for AE-IPF,6,7 based on the rationale that there is an increased inflammatory response in AE-IPF that can be mitigated by immunosuppressive treatment.8 Prior data suggest that the use of corticosteroids for AE-IPF is near ubiquitous, with approximately 80% to 90% of patients receiving corticosteroids when hospitalized, and providers reporting that they primarily use the corticosteroid methylprednisolone for treatment.5,9,10 Dosing of corticosteroids, however, can vary widely, from doses similar to COPD exacerbations (eg, 60-120 mg methylprednisolone) to higher pulse doses (eg, 250-1,000 mg methylprednisolone), with posited varying pharmacokinetic effects.11 Prior observational studies regarding higher- vs lower-dose corticosteroids for AE-IPF have yielded mixed results and been subject to confounding by indication.12, 13, 14, 15 We previously demonstrated wide, idiosyncratic, between-hospital variation in the use of pulse dose corticosteroids for AE-IPF.16 In this study, we sought to leverage the variation in a hospital’s propensity to administer pulse dose methylprednisolone in an instrumental variable analysis emulating a target trial, which is inherently resistant to unmeasured confounding compared with traditional observational methods,17 to examine the association between pulse dose and lower-dose corticosteroids for AE-IPF.
Study Design and Methods
Study Design
We conducted a retrospective cohort study using an instrumental variable design informed by a target trial emulation framework to create a study analogous to a randomized controlled trial. We aimed to study hospitalized patients enrolled in either a pulse or low-dose methylprednisolone arm for AE-IPF treatment. Instrumental variable analysis is a quasiexperimental, econometric method that can leverage idiosyncratic or preference-sensitive decisions in care to mimic randomization and help address the unmeasured confounding by indication that plagues traditional observational study designs.18,19 Target trial emulation frameworks can be adapted for instrumental variable study designs20 to emulate a randomized experiment using causal inference methods from large observational databases by predefining eligibility criteria, treatment strategies, follow-up period, outcome, and analysis plan akin to a trial protocol (e-Table 1).21,22
Data Source and Cohort Selection
We used the Premier Healthcare Database (January 2016-September 2022) containing data representative of approximately 25% of annual US admissions.23 Premier includes all-payer data from > 170 million geographically diverse US inpatient admissions, including demographics, diagnosis codes, medications administered, procedures, and other billing data with a hospital makeup similar to the American Hospital Association database.24 Premier has minimal data missingness, with < 0.01% missing data for key demographic and diagnostic information; specialty pharmacy medications (ie, antifibrotics for IPF) are not captured in the data set.23 We identified a cohort of patients ≥ 50 years of age (IPF is unlikely in adults < 50 years of age)7 with AE-IPF, using validated International Classification of Diseases, 10th Revision, diagnosis codes for either IPF (J84.112) as primary diagnosis25 or acute or chronic respiratory failure or ARDS (J96.x, J80, or R06.03) as primary diagnosis26, 27, 28 with J84.112 as a secondary diagnosis, who received IV methylprednisolone on day 1 or 2 of admission. We have used this algorithm in prior studies.16 Patients were excluded if they (1) had a diagnosis code for an alternate interstitial lung disease or connective tissue disease during admission, (2) had a primary admission diagnosis of congestive heart failure (could mimic AE-IPF) or COPD or asthma (common alternate indications for methylprednisolone), (3) were admitted for major surgery or lung transplant, or (4) were transferred from another facility. If a patient was admitted more than once in the data set, random sampling was used to select 1 hospitalization for inclusion. Codes used for inclusion/exclusion are listed in e-Table 2.
Covariates
To account for baseline differences between patients receiving low-dose and pulse dose methylprednisolone, models were adjusted for patient-level demographics (eg, age, race, ethnicity, insurance), hospital-level factors (eg, bed size, teaching, rural/urban, hospital IPF burden, transplant hospital), acute organ dysfunction and infection on admission,27,29 clinical factors (eg, level of admission [ICU, stepdown], prior chronic oxygen use, mechanical and noninvasive ventilation on admission, prior AE-IPF hospitalizations, do-not-resuscitate status, admission under a pulmonary provider), and chronic comorbidities.30 Codes used to identify covariates are listed in e-Table 3. We used medians (interquartile range [IQR]) and proportions (%) as appropriate to summarize patient- and hospital-level covariables. Standardized mean differences were calculated to quantify difference between groups.
Outcomes
Patients were followed to the end of hospitalization. The primary outcome of interest was a composite outcome of in-hospital death or discharge to hospice. The secondary outcome of interest was discharge home without IMV (identified by the absence of charges for mechanical ventilation on day of discharge). This secondary outcome was selected as a patient-centered outcome representing the ability to discharge home without need for short-term rehabilitation, long-term care, hospice, or long-term ventilatory support (ie, tracheostomy).
Exposure and Instrument
The exposure of interest was receipt of pulse dose methylprednisolone on day 1 or 2 of admission. Pulse dosing was defined as ≥ 250 mg IV methylprednisolone daily based on prior AE-IPF literature.9 The comparator was patients who received lower doses (< 250 mg IV methylprednisolone) starting on day 1 or 2 of admission. Daily doses were calculated from daily billing charges for IV methylprednisolone. We only assessed IV methylprednisolone because this is the most commonly used corticosteroid for AE-IPF,10 and including other corticosteroids may have resulted in increased confounding by indication because choice of corticosteroid is often driven by primary disease being treated (eg, IV hydrocortisone is commonly used for septic shock) or illness severity (eg, patients receiving oral prednisone may have a different illness severity than those receiving IV methylprednisolone).
The chosen instrument was admission to a high pulse dose use hospital (hereafter, high pulse hospital) or low pulse dose use hospital (hereafter, low pulse hospital). To define our instrument, we built hierarchical logistic regression models (including all aforementioned covariables as fixed effects and hospital of admission as a random intercept) with the outcome of pulse dose use; we extracted each hospital’s random effect to estimate each hospital’s predicted rate of pulse dose methylprednisolone use throughout the study period and then ranked hospitals in quintiles of lowest to highest predicted rates. High pulse hospitals were defined as those in the top 2 quintiles of predicted rates, whereas low pulse hospitals were those in the bottom 2 quintiles; the middle (third) quintile of hospitals was removed from analysis to improve the strength of the instrument. Only hospitals with ≥ 10 patients with AE-IPF meeting inclusion were included to improve stability of hospital pulse dose corticosteroid rate estimates.
Instrumental Variable Validity
Steps were taken to test validity of the instrumental variable approach (exclusion, relevance, and exchangeability). A directed acyclic graph was created to evaluate the pathways in which the instrumental variable could affect the outcome of interest to increase certainty that there would be no direct effect on the outcome (exclusion) (e-Fig 1). To further evaluate the assumptions of the exclusion restriction, falsification testing was conducted to assess alternative outcomes unlikely to be affected by steroid treatment but potentially associated with unmeasured confounders linked to the instrument (e-Tables 4-5). Next, an F-statistic for the first step of the model was calculated to assess the strength of the instrument’s correlation with the exposure (relevance) (F-statistic ≥ 100 was a priori considered a strong instrument).31 Instrument relevance was also assessed with a partial R2 to characterize the proportion of variance explained by the addition of the instrument to the model. Then, the distribution of covariates across quintiles of hospital pulse corticosteroids rates was assessed qualitatively (exchangeability) in the primary analysis cohort (e-Table 6) and the subgroup cohorts (e-Tables 7, 8). We calculated standardized mean differences and, a priori, a standardized mean difference < 0.20 was considered to be minimal difference between groups. Additionally, patient characteristics stratified by instrument-assigned pulse exposure with and without the outcome of death/hospice were further explored among subgroups to evaluate balance (e-Tables 9-12).
Statistical Approach
A 2-stage least squares (2SLS) approach was used to assess association of receipt of pulse dose corticosteroids with outcomes. Although outcomes were binary, the commonly used 2SLS approach with linear probability models was selected for straightforward implementation and interpretability within an instrumental variable framework. Prior methodologic work has shown the 2SLS approach performs comparably with alternative instrumental variable methods despite the assumption of normally distributed errors.17 In prespecified subgroup analyses to better understand if the association between pulse dosing and outcomes varied by severity of acute presentation, we first tested for interaction between unit of admission (ICU/intermediate care unit vs general ward units) and exposure with outcome; when significant, we continued with conducting stratified subgroup analyses with relevant outcomes.32 All analyses adjusted for all covariables to account for imbalance between any characteristics across levels of the instrument33; however, given the risk of collider bias when adjusting for covariates in an instrumental variable approach,34 we also performed sensitivity analyses with unadjusted models. To account for within-hospital correlation, we calculated cluster-robust SEs and corresponding CIs, clustering by hospital identifier in the second-stage instrumental variable model. Model results were interpreted as a local average treatment effect (risk difference [RD]) among compliers (ie, patients whose receipt of pulse dose corticosteroids was discretionary and primarily affected by being admitted to a high pulse hospital rather than a low pulse hospital).
Analysis was conducted in R Studio Version 4.1.2 (Posit). Boston University Institutional Review Board deemed this study exempt from Institutional Review Board approval (H-41991).
Results
Patient and Hospital Characteristics
We identified 3,049 patients with AE-IPF at 177 hospitals meeting inclusion criteria in the primary analysis cohort (e-Fig 2). The median age was 75 years (IQR, 68-81) with 1,818 male patients (59.6%), 2,543 White patients (83.4%), 200 Black patients (6.6%), and 276 Hispanic patients (9.1%). There were 1,094 patients who received pulse dose steroids and 1,955 who received low-dose steroids. Patient- and hospital-level characteristics, stratified by patient-level pulse exposure and instrument-assigned exposure (high pulse use vs low pulse use hospitals) are shown in Table 1; after stratification by instrument, almost all patient and hospital characteristics met prespecified criteria for minimal difference between groups. Among high pulse hospitals, the median hospital predicted rate of pulse dosing in the first 2 days of admission was 42.6% (IQR, 36.2%-49.2%); among low pulse hospitals, the median hospital predicted rate of pulse dosing was 14.8% (IQR, 12.5%-18.0%). The median patient-level methylprednisolone dose in the first 2 days of admission at high pulse hospitals was 250 mg (IQR, 125-455 mg) and at low pulse hospitals was 125 mg (IQR, 120-205 mg).
Table 1.
Patient Baseline Characteristics Stratified by Individual Dosing and Hospital-Level Dosing
| Characteristic | Low Dose (n = 1,955) | Pulse Dose (n = 1,094) | SMD | Admitted to Low Use Hospitals (n = 1,550) | Admitted to High Use Hospitals (n = 1,499) | SMD |
|---|---|---|---|---|---|---|
| Age, y | 75.0 (69-81) | 74.0 (67-80) | 0.19 | 75 (69-81) | 74 (67-80) | 0.17 |
| Sex | 0.13 | 0.04 | ||||
| Female | 834 (42.7) | 397 (36.3) | 640 (41.3) | 591 (39.4) | ||
| Male | 1,121 (57.3) | 697 (63.7) | 910 (58.7) | 908 (60.6) | ||
| Race | 0.09 | 0.15 | ||||
| Asian | 35 (1.8) | 17 (1.6) | 39 (2.5) | 13 (0.9) | ||
| Black | 141 (7.2) | 59 (5.4) | 89 (5.7) | 111 (7.4) | ||
| Other | 105 (5.4) | 50 (4.6) | 85 (5.5) | 70 (4.7) | ||
| Unknown | 64 (3.3) | 35 (3.2) | 52 (3.4) | 47 (3.1) | ||
| White | 1,610 (82.4) | 933 (85.3) | 1,285 (82.9) | 1,258 (83.9) | ||
| Ethnicity | 0.19 | 0.24 | ||||
| Hispanic | 214 (10.9) | 62 (5.7) | 192 (12.4) | 89 (5.7) | ||
| Not Hispanic | 1,384 (70.8) | 826 (75.5) | 1,063 (68.6) | 1,147 (76.5) | ||
| Unknown | 357 (18.3) | 206 (18.8) | 295 (19.0) | 268 (17.9) | ||
| Hospital type | 0.01 | 0.09 | ||||
| Rural | 211 (10.8) | 116 (10.6) | 188 (12.1) | 139 (9.3) | ||
| Urban | 1,744 (89.2) | 978 (89.4) | 1,363 (87.9) | 1,360 (90.7) | ||
| Hospital teaching status | 0.05 | 0.01 | ||||
| Nonteaching | 1,024 (52.4) | 599 (54.8) | 820 (52.9) | 803 (53.6) | ||
| Teaching hospital | 931 (47.6) | 495 (45.2) | 730 (47.1) | 696 (46.4) | ||
| Hospital bed size | 0.09 | 0.14 | ||||
| < 100 | 18 (0.9) | 10 (0.9) | 14 (0.9) | 14 (0.9) | ||
| 100-199 | 227 (11.6) | 124 (11.3) | 203 (13.1) | 148 (9.9) | ||
| 200-299 | 346 (17.7) | 198 (18.1) | 271 (17.5) | 273 (18.2) | ||
| 300-399 | 483 (24.7) | 233 (21.3) | 365 (23.5) | 351 (23.4) | ||
| 400-499 | 270 (13.8) | 175 (16.0) | 241 (15.5) | 204 (13.6) | ||
| ≥ 500 | 611 (31.3) | 354 (32.4) | 456 (29.4) | 509 (34.0) | ||
| Hospital region | 0.22 | 0.08 | ||||
| Midwest | 406 (20.8) | 219 (20.0) | 317 (20.5) | 308 (20.5) | ||
| Northeast | 268 (13.7) | 90 (8.2) | 188 (12.1) | 170 (11.3) | ||
| South | 1,090 (55.8) | 622 (56.9) | 847 (54.6) | 865 (57.7) | ||
| West | 191 (9.8) | 163 (14.9) | 198 (12.8) | 156 (10.4) | ||
| Safety-net hospital | 371 (19.0) | 196 (17.9) | 0.03 | 280 (18.1) | 287 (19.1) | 0.03 |
| Hospital IPF burden, % | 0.11 (0.03- 0.33) | 0.12 (0.03- 0.33) | 0.11 | 0.12 (0.10-0.14) | 0.11 (0.08-0.15) | 0.07 |
| Lung transplant hospital | 16 (0.8) | 14 (1.3) | 0.05 | 11 (0.7) | 19 (1.3) | 0.06 |
| Insurance status | 0.11 | 0.04 | ||||
| Medicaid | 78 (4.0) | 56 (4.6) | 68 (4.4) | 66 (4.4) | ||
| Medicare | 1,607 (82.2) | 870 (80.7) | 1,270 (81.9) | 1,207 (80.5) | ||
| Other | 46 (2.4) | 20 (1.8) | 30 (1.9) | 36 (2.4) | ||
| Private | 207 (10.6) | 130 (11.9) | 165 (10.6) | 172 (11.5) | ||
| Uninsured, self-pay, charity | 17 (0.9) | 18 (1.6) | 17 (1.1) | 18 (1.2) | ||
| Level of care | ||||||
| ICU | 257 (13.1) | 222 (20.3) | 0.19 | 241 (15.5) | 238 (15.9) | 0.01 |
| Stepdown unit | 668 (34.2) | 345 (31.5) | 0.06 | 588 (37.9) | 425 (28.4) | 0.04 |
| Primary pulmonary provider | 117 (6.0) | 119 (10.9) | 0.19 | 72 (4.6) | 164 (10.5) | 0.21 |
| Ventilatory support | ||||||
| Invasive ventilation | 82 (4.1) | 84 (7.7) | 0.15 | 80 (5.2) | 86 (5.7) | 0.03 |
| Noninvasive ventilation | 386 (19.7) | 282 (25.8) | 0.14 | 333 (21.5) | 335 (22.3) | 0.02 |
| Chronic oxygen use | 15 (0.8) | 2 (0.2) | 0.09 | 5 (0.3) | 12 (0.8) | 0.06 |
| Acute organ dysfunctiona | ||||||
| Cardiovascular | 99 (5.1) | 49 (4.5) | 0.03 | 77 (5.0) | 71 (4.7) | 0.01 |
| Neurologic | 64 (3.3) | 28 (2.6) | 0.04 | 49 (3.2) | 43 (2.9) | 0.02 |
| Hematologic | 69 (3.5) | 50 (4.6) | 0.05 | 50 (3.2) | 69 (4.6) | 0.07 |
| Hepatic | 4 (0.2) | 2 (0.2) | < 0.01 | 5 (0.3) | 1 (0.1) | 0.05 |
| Renal | 261 (13.4) | 133 (12.2) | 0.04 | 212 (13.7) | 182 (12.1) | 0.02 |
| Vasopressors | 27 (1.4) | 20 (1.8) | 0.04 | 28 (1.8) | 19 (1.3) | 0.04 |
| Infectiona | 1,008 (51.6) | 584 (53.4) | 0.04 | 817 (52.7) | 775 (51.7) | 0.02 |
| Comorbiditiesb | ||||||
| Alcohol misuse | 27 (1.4) | 19 (1.7) | 0.03 | 16 (1.0) | 30 (2.0) | 0.08 |
| Any tumor | 86 (4.4) | 63 (5.8) | 0.06 | 67 (4.3) | 82 (5.5) | 0.05 |
| Cardiac arrythmia | 586 (30.0) | 335 (30.6) | 0.01 | 462 (29.8) | 459 (30.6) | 0.02 |
| Chronic pulmonary disease | 1,361 (69.6) | 707 (64.6) | 0.11 | 1,063 (68.6) | 1,005 (67.0) | 0.03 |
| Coagulopathy | 105 (5.4) | 64 (5.9) | 0.02 | 80 (5.2) | 89 (5.9) | 0.03 |
| Complicated diabetes | 372 (19.0) | 194 (17.7) | 0.03 | 298 (19.2) | 268 (17.9) | 0.03 |
| Congestive heart failure | 777 (39.7) | 386 (35.3) | 0.09 | 612 (39.5) | 551 (36.8) | 0.05 |
| Anemia | 265 (13.6) | 154 (14.1) | 0.02 | 208 (13.4) | 211 (14.1) | 0.02 |
| Dementia | 100 (5.1) | 33 (3.0) | 0.11 | 72 (4.6) | 61 (4.1) | 0.03 |
| Fluid and electrolyte disorders | 616 (31.5) | 367 (33.5) | 0.04 | 475 (30.6) | 508 (33.9) | 0.07 |
| Hypertension | 745 (38.1) | 328 (30.0) | 0.17 | 561 (36.2) | 512 (34.2) | 0.04 |
| Liver disease | 55 (2.8) | 37 (3.4) | 0.03 | 42 (2.7) | 50 (3.3) | 0.04 |
| Metastatic cancer | 25 (1.3) | 15 (1.4) | 0.01 | 23 (1.5) | 17 (1.1) | 0.03 |
| Peripheral vascular disease | 156 (8.0) | 82 (7.5) | 0.02 | 109 (7.0) | 129 (8.6) | 0.06 |
| Psychosis | 307 (15.7) | 176 (16.1) | 0.01 | 215 (13.9) | 268 (17.9) | 0.11 |
| Pulmonary circulation disorders | 708 (36.2) | 383 (35.0) | 0.03 | 565 (36.5) | 526 (35.1) | 0.03 |
| Renal failure | 375 (19.2) | 166 (15.2) | 0.11 | 271 (17.5) | 270 (18.0) | 0.14 |
| Weight loss | 112 (5.7) | 65 (5.9) | 0.01 | 90 (5.8) | 87 (5.8) | < 0.01 |
| Obesity | 272 (13.9) | 191 (17.5) | 0.10 | 207 (13.4) | 256 (17.1) | 0.10 |
| Prior IPF hospitalization | 192 (9.8) | 88 (8.0) | 0.06 | 151 (9.7) | 129 (8.6) | 0.04 |
| Do-not-resuscitate order | 474 (24.2) | 245 (22.4) | 0.04 | 371 (23.9) | 348 (23.2) | 0.02 |
Data are presented as No. (%), median (interquartile range), or as otherwise indicated. All baseline characteristics were measured on hospital day 1 (if charges) or if present on admission (if captured by International Classification of Diseases, 10th Revision, coding). IPF = idiopathic pulmonary fibrosis; SMD = standardized mean difference.
As defined by Angus et al.27
As defined by Sun et al.30
Instrumental Variable Validity
Exclusion was assessed by the directed acyclic graph in e-Figure 1, which was thought to represent an instrument which did not directly affect the outcome other than via the exposure. In falsification testing, there was no statistically significant association between pulse corticosteroid use and the alternate outcomes of hemodialysis initiation, venous thromboembolism, and acute kidney injury, further supporting the exclusion restriction assumption (e-Tables 4, 5). The F-statistic for the primary analysis model of all patients was 705 and for the ICU/intermediate care unit model and ward model was 145 and 291, respectively. This met our a priori relevance criteria for a strong instrument in all models used. Additionally, the partial R2 exceeded 0.10 for all models,35 a threshold suggesting that the instrument met relevance criteria and explained a meaningful proportion of variation in treatment assignment across hospitals (primary model: R2 = 0.19; ICU/intermediate care model: R2 = 0.12; ward model: R2 = 0.21). Finally, exchangeability criteria was met with patient covariates evenly distributed across quintiles of hospital pulse rates other than in the categories of hospital IPF burden, ethnicity, teaching hospital, and bed size.
Model Outcomes
Compared with those who received a low methylprednisolone dose, the RD for in-hospital death/discharge to hospice for those receiving a pulse methylprednisolone dose was 1.2% (95% CI, −6.3% to 8.8%) (Fig 1). The RD for being discharged home without IMV support for those who received a pulse methylprednisolone dose was 5.3% (95% CI, −2.5% to 13.1%) compared with those who received a low methylprednisolone dose. Results of sensitivity analyses with unadjusted models showed consistent findings with an RD of in-hospital death/discharge to hospice of 3.8% (95% CI, −5.1% to 12.7%), and for discharge home without IMV support of 5.5% (95% CI, −5.1% to 16.0%) (e-Table 13).
Figure 1.
Absolute risk differences for outcomes by primary cohort and subgroups where interaction testing was significant. IMV = invasive mechanical ventilation.
In subgroup analysis, we identified 1,069 patients admitted to an ICU or intermediate care unit at 67 hospitals and 1,141 patients admitted to a general ward unit at 77 hospitals. There were 499 patients who received pulse dose steroids and 826 who received low-dose steroids in the ICU or intermediate care unit cohort; there were 453 patients who received pulse dose steroids and 943 who received low-dose steroids in the general ward unit cohort. In models for the risk of hospital death or discharge to hospice, there was a significant interaction between unit of admission and exposure (P for interaction = .0009). The RD for hospital death or hospice discharge associated with receiving a pulse methylprednisolone dose was significantly larger among patients admitted to ICU/intermediate care units than those admitted to a general ward unit (RD, 26.3%; 95% CI, 9.1%-43.5%, vs RD, 0.1%; 95% CI, −10.7% to 10.9%, respectively). There was no significant difference in the risk of being discharged home without IMV support between those admitted to ICU/intermediate care units vs those admitted to a ward unit (P for interaction = .18; RDs not separately quantified as a result). Results were similar in sensitivity analyses (e-Table 13).
Discussion
Leveraging a large national data set, we used an econometric approach to examine outcomes associated with pulse dose methylprednisolone use for patients admitted with AE-IPF. Among patients admitted for AE-IPF whose receipt of pulse dose was discretionary based on hospital practice, we observed no significant difference in the risk of in-hospital death/discharge to hospice or discharge home without IMV with the receipt of pulse dose methylprednisolone. However, among the subgroup of patients admitted to an ICU or intermediate care unit, receipt of pulse dose methylprednisolone was associated with increased risk of in-hospital death/discharge to hospice for patients with AE-IPF.
Although guidelines make a weak recommendation for the use of corticosteroids for AE-IPF, they do not make recommendations as to the optimal dose of corticosteroids.7 In surveys, most providers report using pulse dose methylprednisolone (≥ 250 mg daily) for treatment of AE-IPF, but observational data suggest wider variation in dosing practice,9,16 highlighting the paucity of literature comparing outcomes between pulse dose corticosteroid and low-dose corticosteroid use. Farrand et al13 studied 82 patients at a single US center with AE-IPF and found no difference in in-hospital mortality but reduced overall survival in those who received corticosteroids vs those who did not receive any corticosteroids. A similar single-center study with 117 patients with AE-IPF in Korea found no significant difference in mortality between patients with AE-IPF who received > 1 mg/kg of prednisolone vs those who received < 1 mg/kg of prednisolone,14 whereas a single center study in Spain with 50 patients observed a relationship between higher doses of systemic steroids and in-hospital mortality.15 However, these studies varied by inclusion time periods, illness severity, indications for steroid use, and placebo comparator groups; they may have also been vulnerable to confounding by indication. Most recently, Hyung et al12 published a retrospective, propensity-matched analysis of 238 patients with AE-IPF treated at a single center in South Korea with either pulse (methylprednisolone of ≥ 250 mg/d) or non-pulse dose corticosteroids, and found no difference in either 3-month or 1-year mortality between groups. Our study substantiates these findings with a larger, multicenter cohort using strict enrollment criteria and an instrumental variable approach to further address residual confounding in observational data.
Of note, when we looked at the subgroup of patients admitted to an ICU or step-down unit, we did see a significantly greater risk of death with receipt of pulse dose methylprednisolone, compared with the RD associated with pulse dose methylprednisolone observed in patients admitted to the general ward. It is possible that patients admitted to an ICU or intermediate care unit have more severe presentations because of more advanced underlying fibrotic lung disease. There is belief that the immunosuppression from corticosteroids may exacerbate disturbances in the lung microbiome in AE-IPF and worsen outcomes,36 and ICU/intermediate care patients with more advanced fibrosis may gain less benefit from antiinflammatory treatment of acute exacerbations while remaining vulnerable to the infectious complications of pulse dose corticosteroids. It is also possible that, despite a high F-statistic, our ICU/intermediate care model had residual confounding, and our findings warrant further investigation, ideally with a randomized trial investigating pulse vs low-dose corticosteroids among the sickest patients with AE-IPF.
Our study has limitations. As in any study using administrative data, misclassification of AE-IPF is possible, particularly given unavailability of admission CT imaging findings. Given this risk, we followed validated algorithms for identifying AE-IPF25 and comorbidities27, 28, 29, 30,37 and further excluded confounding conditions from both diagnostic and treatment standpoints to improve specificity. We adjusted for severity of baseline IPF by accounting for chronic oxygen use and prior hospitalizations for AE-IPF; however, we did not have IPF diagnosis date, prior imaging, pulmonary function testing, outpatient medication use, or degree of hypoxemia available to fully account for baseline IPF severity. Because of the lack of billing charges for specialty medications, we could not assess the impact of antifibrotic use on outcomes. Furthermore, even in a large, multiyear national database, our subgroup analyses had small cohort sizes, which may have limited the power of our instrumental variable analyses.31 Relatedly, SEs for instrumental variable approaches are known to be substantially larger when compared with ordinary regression and a 2SLS model requires assumptions when used with a dichotomous outcome.17 A 2SLS model also does not allow for time-to-event analyses.38 Despite these limitations, in the absence of randomized data, our study adds to the literature on steroid dosing in AE-IPF by using causal inference methodology to address the potential for unmeasured confounding in observational datasets.
Intepretation
In an instrumental variable analysis, we observed no significant benefit or harm with the use of pulse dose methylprednisolone for AE-IPF among all patients. However, among the subgroup of patients admitted to an ICU or intermediate care unit, receipt of pulse dose methylprednisolone was associated with an increased risk of in-hospital death/discharge to hospice. This subgroup analysis is hypothesis-generating and highlights the need for further studies to evaluate corticosteroid dosing practices in the sickest patients with AE-IPF, ideally in a prospective, randomized trial.
Funding/Support
D. A. S. was supported by 1F32HL168959-01 from NIH/NHLBI. R. S. W. is supported in part by resources from the VA Boston Healthcare System.
Financial/Nonfinancial Disclosures
None declared.
Acknowledgments
Author contributions: D.A.S. is the guarantor of the content of the manuscript, including the data and analysis. D. A. S., N. A. B., A. J. W., and A. C. L. were involved in the planning of this study. D. A. S., N. A. B., S. R. J., A. J. W., and A. C. L. were involved in the data collection and data analysis of this study. D. A. S., N. A. B., S. R. J., M.-L. D., R. S. W., K. C. W., F. J. H., K.-D. A., A. J. W., and A. C. L. were involved in writing the manuscript for this study.
Disclaimer: The views expressed are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the US Government.
Role of sponsors: The sponsor had no role in the design of the study, the collection and analysis of the data, or the preparation of the manuscript.
Additional information: The e-Figures and e-Tables are available online under “Supplementary Data.”
Supplementary Data
References
- 1.Lederer D.J., Martinez F.J. Idiopathic pulmonary fibrosis. N Engl J Med. 2018;378(19):1811–1823. doi: 10.1056/NEJMra1705751. [DOI] [PubMed] [Google Scholar]
- 2.Collard H.R., Ryerson C.J., Corte T.J., et al. Acute exacerbation of idiopathic pulmonary fibrosis. An International Working Group Report. Am J Respir Crit Care Med. 2016;194(3):265–275. doi: 10.1164/rccm.201604-0801CI. [DOI] [PubMed] [Google Scholar]
- 3.Natsuizaka M., Chiba H., Kuronuma K., et al. Epidemiologic survey of Japanese patients with idiopathic pulmonary fibrosis and investigation of ethnic differences. Am J Respir Crit Care Med. 2014;190(7):773–779. doi: 10.1164/rccm.201403-0566OC. [DOI] [PubMed] [Google Scholar]
- 4.Durheim M.T., Judy J., Bender S., et al. A retrospective study of in-hospital mortality in patients with idiopathic pulmonary fibrosis between 2015 and 2018. Medicine (Baltimore) 2020;99(47) doi: 10.1097/MD.0000000000023143. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Shankar D.A., Walkey A.J., Hawkins F.J., Bosch N.A., Peterson D., Law A.C. Hospital-level variation in practices and outcomes for patients with severe acute exacerbations of idiopathic pulmonary fibrosis: a retrospective multicentre cohort study. BMJ Open Respir Res. 2023;10(1) doi: 10.1136/bmjresp-2022-001593. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Raghu G., Anstrom K.J., King T.E., Jr., Lasky J.A., Martinez F.J. Prednisone, azathioprine, and N-acetylcysteine for pulmonary fibrosis. N Engl J Med. 2012;366(21):1968–1977. doi: 10.1056/NEJMoa1113354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Raghu G., Collard H.R., Egan J.J., et al. An Official ATS/ERS/JRS/ALAT statement: idiopathic pulmonary fibrosis: evidence-based guidelines for diagnosis and management. Am J Respir Crit Care Med. 2011;183(6):788–824. doi: 10.1164/rccm.2009-040GL. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Bhatti H., Girdhar A., Usman F., Cury J., Bajwa A. Approach to acute exacerbation of idiopathic pulmonary fibrosis. Ann Thorac Med. 2013;8(2):71–77. doi: 10.4103/1817-1737.109815. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Kreuter M., Polke M., Walsh S.L.F., et al. Acute exacerbation of idiopathic pulmonary fibrosis: international survey and call for harmonisation. Eur Respir J. 2020;55(4) doi: 10.1183/13993003.01760-2019. [DOI] [PubMed] [Google Scholar]
- 10.Polke M., Kondoh Y., Wijsenbeek M., et al. Management of acute exacerbation of idiopathic pulmonary fibrosis in specialised and non-specialised ILD centres around the world. Front Med (Lausanne) 2021;8 doi: 10.3389/fmed.2021.699644. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Edel Y., Avni T., Shepshelovich D., et al. The safety of pulse corticosteroid therapy- Systematic review and meta-analysis. Semin Arthritis Rheum. 2020;50(3):534–545. doi: 10.1016/j.semarthrit.2019.11.006. [DOI] [PubMed] [Google Scholar]
- 12.Hyung K., Lee J.H., Kim J.Y., Choi S.M., Park J. Pulse versus non-pulse corticosteroid therapy in patients with acute exacerbation of idiopathic pulmonary fibrosis. Respirology. 2024;29(3):235–242. doi: 10.1111/resp.14643. [DOI] [PubMed] [Google Scholar]
- 13.Farrand E., Vittinghoff E., Ley B., Butte A.J., Collard H.R. Corticosteroid use is not associated with improved outcomes in acute exacerbation of IPF. Respirology. 2020;25(6):629–635. doi: 10.1111/resp.13753. [DOI] [PubMed] [Google Scholar]
- 14.Jang H.J., Yong S.H., Leem A.Y., et al. Corticosteroid responsiveness in patients with acute exacerbation of interstitial lung disease admitted to the emergency department. Sci Rep. 2021;11:5762. doi: 10.1038/s41598-021-85539-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Cuerpo S., Moisés J., Hernández-González F., et al. Acute exacerbations of idiopathic pulmonary fibrosis: Does clinical stratification or steroid treatment matter? Chron Respir Dis. 2019;16 doi: 10.1177/1479973119869334. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Shankar D.A., Hawkins F.J., Alysandratos K.D., et al. Practice patterns of pulse dose corticosteroid use for patients hospitalized with acute exacerbations of idiopathic pulmonary fibrosis in the United States. Chest Pulm. 2024;3(3) doi: 10.1016/j.chpulm.2024.100126. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Rassen J.A., Schneeweiss S., Glynn R.J., Mittleman M.A., Brookhart M.A. Instrumental variable analysis for estimation of treatment effects with dichotomous outcomes. Am J Epidemiol. 2009;169(3):273–284. doi: 10.1093/aje/kwn299. [DOI] [PubMed] [Google Scholar]
- 18.Iwashyna T.J., Kennedy E.H. Instrumental variable analyses. exploiting natural randomness to understand causal mechanisms. Ann Am Thorac Soc. 2013;10(3):255. doi: 10.1513/AnnalsATS.201303-054FR. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Maciejewski M.L., Dowd B.E., Norton E.C. Instrumental variables and heterogeneous treatment effects. JAMA. 2022;327(12):1177–1178. doi: 10.1001/jama.2022.2505. [DOI] [PubMed] [Google Scholar]
- 20.Swanson S.A. Instrumental variable analyses in pharmacoepidemiology: What target trials do we emulate? Curr Epidemiol Rep. 2017;4(4):281–287. doi: 10.1007/s40471-017-0120-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hernán M.A., Robins J.M. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758–764. doi: 10.1093/aje/kwv254. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Usman M.S., Pitt B., Butler J. Target trial emulations: bridging the gap between clinical trial and real-world data. Eur J Heart Fail. 2021;23(10):1708–1711. doi: 10.1002/ejhf.2331. [DOI] [PubMed] [Google Scholar]
- 23.Premier Applied Sciences Premier Healthcare Database: data that informs and performs. https://offers.premierinc.com/Premier-Healthcare-Database-Download.html
- 24.AHA Data & Insights. https://www.ahadata.com/
- 25.Durheim M.T., Judy J., Bender S., et al. In-hospital mortality in patients with idiopathic pulmonary fibrosis: a US cohort study. Lung. 2019;197(6):699–707. doi: 10.1007/s00408-019-00270-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Centers for Medicare and Medicaid Services ICD-10-CM official guidelines for coding and reporting. https://www.cms.gov/files/document/fy-2024-icd-10-cm-coding-guidelines-updated-02/01/2024.pdf
- 27.Angus D.C., Linde-Zwirble W.T., Lidicker J., Clermont G., Carcillo J., Pinsky M.R. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Crit Care Med. 2001;29(7):1303–1310. doi: 10.1097/00003246-200107000-00002. [DOI] [PubMed] [Google Scholar]
- 28.Dombrovskiy V.Y., Martin A.A., Sunderram J., Paz H.L. Rapid increase in hospitalization and mortality rates for severe sepsis in the United States: a trend analysis from 1993 to 2003∗. Crit Care Med. 2007;35(5):1244. doi: 10.1097/01.CCM.0000261890.41311.E9. [DOI] [PubMed] [Google Scholar]
- 29.Bosch N.A., Law A.C., Rucci J.M., Peterson D., Walkey A.J. Predictive validity of the sequential organ failure assessment score versus claims-based scores among critically ill patients. Ann Am Thorac Soc. 2022;19(6):1072–1076. doi: 10.1513/AnnalsATS.202111-1251RL. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Sun J.W., Rogers J.R., Her Q., et al. Adaptation and validation of the combined comorbidity score for ICD-10-CM. Med Care. 2017;55(12):1046–1051. doi: 10.1097/MLR.0000000000000824. [DOI] [PubMed] [Google Scholar]
- 31.Moler-Zapata S., Grieve R., Basu A., O’Neill S. How does a local instrumental variable method perform across settings with instruments of differing strengths? A simulation study and an evaluation of emergency surgery. Health Econ. 2023;32(9):2113–2126. doi: 10.1002/hec.4719. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Altman D.G., Bland J.M. Interaction revisited: the difference between two estimates. BMJ. 2003;326(7382):219. doi: 10.1136/bmj.326.7382.219. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Garabedian L.F., Chu P., Toh S., Zaslavsky A.M., Soumerai S.B. Potential bias of instrumental variable analyses for observational comparative effectiveness research. Ann Intern Med. 2014;161(2):131–138. doi: 10.7326/M13-1887. [DOI] [PubMed] [Google Scholar]
- 34.Coscia C., Gill D., Benítez R., Pérez T., Malats N., Burgess S. Avoiding collider bias in Mendelian randomization when performing stratified analyses. Eur J Epidemiol. 2022;37(7):671–682. doi: 10.1007/s10654-022-00879-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Rassen J.A., Brookhart M.A., Glynn R.J., Mittleman M.A., Schneeweiss S. Instrumental variables II: instrumental variable application—in 25 variations, the physician prescribing preference generally was strong and reduced covariate imbalance. J Clin Epidemiol. 2009;62(12):1233–1241. doi: 10.1016/j.jclinepi.2008.12.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Papiris S.A., Kagouridis K., Kolilekas L., et al. The new idiopathic pulmonary fibrosis acute exacerbations document: one step ahead but still suspended in the air. Am J Respir Crit Care Med. 2017;195(2):267–269. doi: 10.1164/rccm.201607-1426LE. [DOI] [PubMed] [Google Scholar]
- 37.Walkey A.J., Barnato A.E., Rinne S.T., et al. Hospital variation in do-not-resuscitate orders and end-of-life healthcare use in the United States. Ann Am Thorac Soc. 2017;14(9):1485–1489. doi: 10.1513/AnnalsATS.201704-303RL. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Martinussen T., Vansteelandt S., Tchetgen Tchetgen E.J., Zucker D.M. Instrumental variables estimation of exposure effects on a time-to-event endpoint using structural cumulative survival models. Biometrics. 2017;73(4):1140–1149. doi: 10.1111/biom.12699. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.