Decline in Prolonged Acute Mechanical Ventilation, 2011–2019

To the Editor:

Estimates from the early 2000s projected that Medicare would face substantial strain owing to disproportionately rapid growth and resource use by older patients requiring “prolonged acute mechanical ventilation” (PAMV) (i.e., mechanical ventilation [MV] for ⩾96 h) (1–4). The number of adults >65 years old receiving PAMV was projected to steadily rise and lead to overwhelming ICU and postacute facility use by 2020 (3, 4). Furthermore, one-third of patients receiving PAMV go on to receive prolonged mechanical ventilation (PMV) (i.e., MV for more than14–21 d), a state of persistent critical illness with high morbidity and mortality (5, 6). However, actual trends in older adults receiving PAMV in the last decade are unclear. We conducted an epidemiologic study among Medicare beneficiaries to determine recent trends in PAMV, decisions to limit invasive treatments, and patient outcomes.

Methods

We used the Medicare Provider Analysis and Review (MedPAR) and Master Beneficiary Summary Files from the Centers for Medicare and Medicaid Services to identify fee-for-service beneficiaries years (>65) old admitted to an ICU between January 1, 2011, and December 31, 2019, with PAMV (2, 3) (previously defined [2–4], International Classification of Diseases, Ninth Edition [ICD-9] procedure code 96.72, “MV ⩾ 96 hours”, and its ICD-10 equivalent [7], 5A1955Z). Among patients with multiple eligible admissions, one admission per patient was randomly selected for inclusion. The primary outcomes were annual population-standardized rates of patients receiving PAMV and the annual proportion of patients receiving PAMV out of total patients receiving any MV (PAMV/MV; ICD-9 96.7×; ICD-10 5A19[3–5]5Z). Among patients receiving PAMV, we also determined annual rates of tracheostomy (ICD-9 31.1, 31.2×; ICD-10 0B11[0–4]F4, 0B11[0–4]Z4), median length of stay, discharge destination, do-not-resuscitate orders (DNR, ICD-9 V49.86; ICD-10 Z66), and palliative care delivery (ICD-9 V66.7; ICD10 Z51.5). Among patients receiving PAMV discharged before December 31, 2018 (i.e., ⩾1 year of posthospitalization follow-up available), we also determined annual 1-year mortality and median institution-free days (i.e., days not in an acute care hospital, long-term acute care hospital, or skilled nursing facility) (8) within 365 days of admission.

We tested for significant trends with Cochrane-Armitage (binary), Cochran-Mantel-Haenszel (categorical), and Jonckheere-Terpstra (continuous outcomes) tests. To assess the association between palliative care delivery and DNR orders and 1) tracheostomy rates and 2) discharge to long-term acute care hospital or skilled nursing facility, we used multivariable hierarchical logistic regression with DNR orders or palliative care as exposures of interest, hospital of admission as a random intercept, and hospital- and patient-level covariables as fixed effects (hospital size, region, rural/urban status, teaching status, safety-net status, profit/ownership status; patient sex, age, race, social vulnerability index [9], Medicaid dual-eligibility, surgical status [10], comorbidities [11], and presence of acute organ dysfunction [cardiovascular, renal, hepatic, hematologic, and neurologic failure]) (12). Statistical testing was two-tailed, with α = 0.05, using SAS version 9.4 (SAS Institute, Inc). This study was deemed exempt from review by Beth Israel Deaconess Medical Center (#2022P000105) and Boston University Institutional Review Boards (#H-41346).

Results

Between 2011 and 2019, 1,823,659 mechanically ventilated patients met inclusion criteria; 646,677 patients (35.5%) received PAMV (47.6% female; mean age, 76.7 ± 7.4 yr). Characteristics of patients in select years is shown in Table 1.

Table 1.

Characteristics of Patients Receiving Prolonged Acute Mechanical Ventilation in Select Years

Relevant Covariables	2011 (N = 78,504)	2013 (N = 75,006)	2015 (N = 74,180)	2017 (N = 69,534)	2019 (N = 60,625)	Total (N = 641,782)
	Female, %	50.0	49.4	46.6	46.0	45.1	4760.0
Age, mean ± SD	77.5 ± 7.5	77.1 ± 7.5	76.6 ± 7.5	76.2 ± 7.4	76.1 ± 7.3	76.7 ± 7.4
Race, %
White non-Hispanic	78.5	77.9	78.3	77.9	76.8	77.9
Black non-Hispanic	14.5	14.6	14.3	14.3	14.7	14.5
Other	7.0	7.6	7.4	7.9	8.5	7.6
Social Vulnerability Index, mean ± SD (N)*	0.54 ± 0.29	0.54 ± 0.29	0.53 ± 0.29	0.53 ± 0.29	0.53 ± 0.29	0.53 ± 0.29
Medicaid/Medicare dual eligibility	25.6	25.5	23.8	24.0	23.6	24.6
Hospital size, %†
Small (1–200 beds)	15.7	15.6	15.1	14.9	14.2	15.2
Medium (201–400)	37.2	36.7	36.2	36.4	35.8	36.4
Large (>400)	47.1	47.7	48.7	48.7	50.0	48.4
Hospital region, %‡
Northeast	20.1	19.7	20.1	19.0	19.2	19.5
Midwest	13.1	12.9	13.5	13.8	13.8	13.4
South	50.9	51.4	50.6	50.6	50.0	50.8
West	15.9	16.0	15.9	16.7	17.0	16.2
Rural†	0.3	0.3	0.3	0.4	0.4	0.3
Teaching hospital†	76.8	76.4	77.1	77.2	78.0	77.0
Safety net hospital§	32.1	32.1	31.4	31.9	32.1	31.9
Ownership†
For profit	17.6	17.5	17.1	17.5	16.3	17.3
Private nonprofit	71.8	71.7	71.4	71.1	72.0	71.5
Public	10.6	10.8	11.5	11.5	11.7	11.2
Surgical patient	47.3	46.3	44.5	39.5	39.0	43.6
Select comorbidities
Alzheimer's disease and related disorders or senile dementia	44.4	43.1	41.9	48.3	47.0	45.0
Breast, colorectal, endometrial, prostate cancer	14.8	14.5	14.2	14.0	13.5	14.2
Lung cancer	5.9	5.7	5.4	5.3	5.6	5.6
Congestive heart failure	73.7	72.3	69.6	69.1	68.6	70.6
Chronic kidney disease	73.1	75.4	77.3	81.8	82.6	78.0
Chronic obstructive pulmonary disease and bronchiectasis	54.8	53.7	52.1	51.7	49.1	52.2
Diabetes	56.5	56.8	54.7	54.9	54.6	55.5
Hypertension	90.9	91.6	91.8	92.2	92.2	91.8
Ischemic heart disease	73.4	73.1	70.9	71.1	70.7	71.9
Stroke/transient ischemic attack	26.4	26.1	26.6	27.7	29.2	27.1
Acute organ dysfunction
Cardiovascular	46.4	50.2	54.2	57.8	59.7	53.4
Neurologic	29.9	34.3	38.7	43.5	47.9	38.5
Hematologic	19.2	21.2	22.2	23.9	22.2	21.7
Hepatic	5.6	6.7	7.5	8.3	9.5	7.4
Renal	54.7	57.0	60.3	63.1	64.6	59.8
Most common principal diagnosis (International Classification of Diseases 9th or 10th revision)
First most common diagnosis; %	038.9, sepsis; 18.1	038.9, sepsis; 21.6	038.9 + A41.9, sepsis; 24.1	A41.9, sepsis; 25.6	A41.9, sepsis; 24.4	038.9 + A41.9, sepsis; 22.8
Second most common diagnosis; %	518.81, acute respiratory failure; 11.5	518.81, acute respiratory failure; 10.0	518.81 + J96.01, acute respiratory failure; 7.3	J96.01, acute respiratory failure; 5.1	J96.01, acute respiratory failure; 5.0	518.81 + J96.01, acute respiratory failure; 7.8

^*Social Vulnerability Index for each patient’s census tract is based on 15 social factors, including poverty, lack of vehicle access, and crowded housing. A census tract’s index is a percentile ranking and reflects the proportion of tracts in the country that are equal to or lower in terms of social vulnerability. Nonmissing (N) = 77,804 (2011), 74,350 (2013), 73,494 (2015), 68,894 (2017), 60,048 (2019), and 635,868 (2011–2019).

^† Hospital size, rural, teaching, ownership statuses; nonmissing (N) = 76,490 (2011), 73,562 (2013), 72,934 (2015), 68,502 (2017), 59,684 (2019), and 630,185 (2011–2019).

^‡ Hospital region; nonmissing (N) = 76,401 (2011), 73,473 (2013), 72,823 (2015), 68,414 (2017), 59,616 (2019), and 629,353 (2011–2019).

^§ Safety net hospital, defined as hospitals with a Medicare Disproportionate Share Index in the top quartile per U.S. region; nomissing (N) = 73,926 (2011), 71,801 (2013), 70,793 (2015), 66,484 (2017), 58,054 (2019), and 610,601 (2011–2019).

Population-standardized incidence of PAMV decreased from 189 of 100,000 adults aged >65 years to 112 of 100,000 adults (78,504 to 60,625 individuals; P < 0.001), concurrent with both decreases in total MV (210,791 to 189,414) and the proportion of PAMV/MV (37% to 32%) (Figure 1). Among patients receiving PAMV, tracheostomies declined from 21% to 17.5% (P < 0.001). Median ICU and hospital length of stay declined from 13 to 12 days (P < 0.001) and 16 to 15 days (P < 0.001), respectively. Discharge to long-term acute care hospitals was stable at ∼17%; discharge to skilled nursing facilities decreased from 22% to 20%, and discharge to hospice increased from 7% to 10% (P < 0.001). DNR orders increased (13.5% to 35.1%; P < 0.001) with concurrent increases in palliative care delivery (15.2% to 30.5%; P < 0.001). One-year mortality remained stable (∼72% [P = 0.19]), as did median institution-free days out of 365 (0 days; interquartile ranges across years, 0 [197–212]). Palliative care delivery was associated with decreased odds of tracheostomy (adjusted odds ratio, 0.42; 95% confidence interval, 0.41–0.43) and transfer to long-term acute care hospitals/skilled nursing facilities (0.09 [0.09–0.09]). Similarly, DNR orders were associated with decreased odds of tracheostomy (0.41 [0.40–0.42]) and transfer to long-term acute care hospitals/skilled nursing facilities (0.20 [0.20–0.20]).

Figure 1.

Trends in prolonged acute mechanical ventilation (PAMV). (A) Population-standardized incidence of PAMV (orange) and proportion of PAMV out of MV (blue) both declined over time. (B) Tracheostomy rates (blue) declined among patients receiving PAMV, whereas DNR rates (orange) and palliative care delivery (gray) increased. (C) Discharge to long-term acute care hospitals (blue) remained stable over time, whereas discharge to skilled nursing facilities (orange) and discharge home (gray) declined; discharge to hospice (yellow) increased. (D) 1-year mortality remained stable.

Discussion

Deviating from earlier trends and projections, the incidence of PAMV declined among older adults from 2011 to 2019. Among patients receiving PAMV, tracheostomy placement declined, whereas DNR rates, palliative care use, and hospice use all increased. Despite increasing limitations on care, mortality and days alive out of institutions were stable.

Our study updates prior work by Zilberberg and colleagues, who used data from the National Inpatient Sample from 2000 to 2008. They found disproportionately rapid growth in adults requiring PAMV and projected an incidence of more than 300,000 in adults aged >65 years by 2020, almost five times higher than incidence we observed in 2019 (2, 4). Several factors may explain the unexpected reversal in PAMV trends. First, invasive mechanical ventilation (of any duration) decreased among Medicare beneficiaries, a phenomenon in line with trends seen in other populations (13) and partially attributed to increased use of noninvasive ventilation (14), changing sedation practices (15), and increased attention to spontaneous awakening and breathing trials (16). Second, the increasing use of palliative care and DNR orders observed was strongly associated with decreased odds of tracheostomy and postacute facility use. Whereas prior research had not shown that the availability of hospital-based palliative care programs was associated with a change in tracheostomy rates among ICU patients (17), our findings in both patient- and national-level analyses suggest that the rapid growth of palliative care during the last decade may have influenced downstream care decisions for patients with PAMV, without effects on 1-year mortality or institution-free days. It remains unclear how changes in PAMV incidence, palliative care use, or DNR orders might affect the subset of patients who go on to receive PMV.

Our study has limitations. First, the data are based on administrative coding and may be vulnerable to potential misclassification. However, for the sake of comparison with prior trends, we used the same ICD-9 (and crosswalked, equivalent ICD-10) codes as prior studies. Because ICD codes in Medicare may slightly overestimate the proportion of patients receiving PAMV (18), the true incidence of patients receiving PAMV may be even lower than observed trends. Similarly, given limitations in ICD coding for DNR orders and palliative care delivery (both with lower sensitivity but higher specificity) (19, 20), our data may underestimate incidence, but can still be interpreted as representing relative trends over time. Second, translation between ICD-9 and ICD-10 may lead to discontinuities in trend data; however, we used previously published ICD-9/ICD-10 algorithms (21), and verified continuity of codes from Centers for Medicare and Medicaid General Equivalence Mapping across 2014–2016 (7). Third, because prior literature estimated significant Medicare strain due to PAMV among patients older than 65 years, we limited our cohort to Medicare beneficiaries. Fourth, although we found notable trend changes in DNR rates, palliative care, and hospice use in our administrative data, the underlying drivers of these shifts—and the potential causal relationship between these practices and PAMV incidence/outcomes—are unclear. Further studies are needed to explore trends in incidence of PAMV in adults younger than 65, how trends may have changed during the COVID-19 pandemic, and the clinical, cultural, and psychologic factors underlying our findings.

In summary, in contrast to earlier trends and projections of increasing incidence of PAMV among older patients, our updated analysis found declines in incidence of PAMV, as well as a strong association between DNR orders and palliative care delivery with decreased odds of tracheostomy and postacute care facility use. Our findings may reflect shifts in patient goals of care over time.