Ascertaining Death Events in a Pragmatic Clinical Trial: Insights from the TRANSFORM-HF Trial

Abstract

Background:

Death ascertainment can be challenging for pragmatic clinical trials that limit site follow-up activities to usual clinical care.

Methods and Results:

We used blinded aggregate data from the ongoing TRANSFORM-HF pragmatic clinical trial in patients with heart failure to evaluate the agreement between centralized call center death event identification and the United States National Death Index (NDI). Of 2284 total patients randomized through April 12, 2021, 1480 were randomized in 2018–2019 and 804 in 2020–2021. The call center identified 416 total death events (177 in 2018–2019 and 239 in 2020–2021). The NDI 2018–2019 final file identified 178 death events, 165 of which were also identified by the call center. The study’s inter-rater reliability metric (Cohen’s kappa coefficient, 0.920; 95% CI, 0.889 to 0.951) demonstrates a high level of agreement. The time between a death event and its identification was less for the call center (median, 47 days; interquartile range (IQR), 11 to 103 days) than for the NDI (median, 270 days; IQR, 186 to 391 days).

Conclusion:

There is substantial agreement between deaths identified by a centralized call center and the NDI. However, the time between a death event and its identification is significantly less for the call center.

Introduction

Explanatory clinical trials typically rely upon site study coordinators to identify patient death events (trigger) and forward verifying documentation to a central clinical events committee (CEC) for adjudication (confirmation) (1). This approach is not feasible for pragmatic clinical trials that limit site follow-up to usual clinical care (2).

The United States (US) National Death Index (NDI) is the most complete US death data source for health statistical reporting and analysis (3). However, there can be nearly a two-year time lag before a death event is availability in the NDI. The ToRsemide comparison with furoSemide FOR Management of Heart Failure (TRANSFORM-HF) pragmatic clinical trial utilizes a follow-up procedure that combines a centralized call center with NDI reports (4). We evaluated the agreement between these death event identification methods using blinded, aggregate information from the ongoing TRANSFORM-HF pragmatic clinical trial. We also estimated differences in the time between a death event and its identification by these methods.

Methods

Study Design

TRANSFORM-HF is a pragmatic randomized clinical trial that compares two loop diuretic therapies (torsemide and furosemide) in patients admitted to a hospital for new or worsening heart failure with outpatient plans for daily, long-term loop diuretic administration (5). The primary endpoint is all-cause mortality. Recruitment occurs at 60 US sites with the first patient enrolled on June 18, 2018. Sites recruit and enroll patients, randomize them to furosemide or torsemide outpatient prescriptions, and collect index hospitalization data. All post-discharge patient follow-up is conducted by the Duke Clinical Research Institute’s (DCRI) Call Center. All patients gave informed consent and their institutional review boards approved the TRANSFORM-HF protocol.

Death Event Ascertainment

TRANSFORM-HF utilizes two, independent methods to obtain death event data: the DCRI’s Call Center and NDI reports (4). All death events require both a trigger (notification of death event) and confirmation (source confirming the death). The DCRI’s Call Center contacts patients at 30days, 6 months and 12-months follow-up, with the first 1500 patients having up to 30-months follow-up (with contact every 6 months). This patient contact schedule creates a 0-to-6 months lag between a patient death and the DCRI Call Center’s ascertainment of the death event. NDI’s preliminary report (approximately 90% of deaths for a year) typically is available in February of the subsequent year and the final report (all deaths for a year) typically is available the following October (e.g., deaths from 2020 are preliminarily provided in February of 2021, with final death data in October of 2021). This creates a 2-to-13 months lag for NDI preliminary death reporting and an 10–21 months lag for final death reporting.

Enrolling TRANSFORM-HF sites identify index hospitalization deaths (trigger) and forward discharge summaries (confirmation) to the DCRI Call Center. The DCRI Call Center identifies patient deaths during the follow-up period. If a patient cannot be reached for an interval interview, the DCRI Call Center uses a protocol to determine patient vital status. The protocol’s death event triggers include: patient-identified proxies (individuals), online searches, and patient-identified hospital and clinic medical records. Death event confirmation methods include: a second proxy confirming the patient’s death, an obituary or grave marker, medical records documenting the patient’s death, and NDI report. The NDI receives annual death information from 57 US jurisdictions (1). Researchers access the NDI by submitting study subject records in a standard format (3). The NDI returns records with a class and an agreement score. We used the NDI definition of True Matches.

Data Sources

The TRANSFORM-HF patient contact form and case report form are used by the DCRI Call Center and NDI death ascertainment processes. Patients complete the patient contact form during their index hospitalization. This form contains detailed contact information for the patient, their spouse/significant other, two friends or relatives not living with the patient, two hospitals the patient would attend, and two health care providers or clinics where the patient would receive post-discharge care. This form also contains information used in NDI searches: patient’s legal name (last name, first name, and middle initial), Social Security Number, state of birth and father’s surname). The TRANSFORM-HF case report form includes four variables used in NDI searches: the patient’s sex, race, state of residence, and marital status.

Study Outcomes and Analyses

This study’s primary outcome is the inter-rater reliability between DCRI Call Center and NDI ascertained 2018–2019 death events (assessed using Cohen’s kappa). Secondary outcomes include the number of death events first identified by the DCRI Call Center and NDI searches and the time interval between a death event and its identification by these methods.

Results

Table 1 shows patient death events grouped by year of enrollment (2018–2019 and 2020–2021) and data source (DCRI Call Center triggers and NDI True Matches). Of 2284 total patients randomized through April 12, 2021, 1480 were randomized in 2018–2019 and 804 in 2020–2021. 1479 were included in the NDI 2018 −2019 Final File search (178 deaths and 1301 non deaths). The remaining 805 patients will be included in subsequent NDI searches. The DCRI Call Center triggered a total of 416 deaths (177 in 2018–2019 and 239 in 2020–2021). Of patients enrolled in 2018–2019, the Call Center and NDI both identified the same 165 deaths occurring in 2018–2019, the NDI identified 13 deaths not identified by the Call Center and the Call Center identified 12 deaths not identified in the NDI 2019 final file. The inter-rater reliability metric (Cohen’s kappa coefficient, 0.920; 95% CI, 0.889 to 0.951) demonstrates a high level of agreement between Call Center and NDI ascertainment of 2018–2019 deaths. The Call Center also identified 169 patients randomized in 2018–2019 who died in 2020–2021 and 70 patients who were randomized and died in 2020–2021. These 2020–2021 deaths will be included in future NDI searches.

Table 1:

Death Events Records Agreement

Call Center Triggers	NDI Death Events – 2018–2019 Final File
	Yes	No	Not Searched	Total
Total Enrolled
Died 2018/2019	165	12	0	177
Died 2020/2021	NA	169	70	239
Not Died	13	1120	735	1868
Total	178	1301	805	2284

Not Searched = 804 patients enrolled in 2020–2021 and 1 patient enrolled in 2018–2019 NA = not included in NDI 2019 final file search.

Table 2 shows triggers and confirmation results for the 429 TRANSFORM-HF death events identified in Table 1. In all, 340 (79%) were confirmed and 89 (21%) have not yet been confirmed. Obituaries and grave markers were the primary confirmation method (n=245), followed by NDI searches (n=70) and medical record searches (n=25). There were 13 True Matches identified by NDI search that were not triggered by the Call Center. Of these, 10 patients verbally modified their informed consent to limit follow-up contact (7 withdrew permission for telephone contact or medical records review and 3 only permitted medical records review). Additionally, 3 patients and their proxies could not be contacted by the Call Center. Of 12 Call Center patient death triggers not returned as NDI True Matches, 8 were verified by obituaries/grave markers, 1 was verified by the enrolling site’s medical records, and 3 were triggered by a proxy but not verified. These 3 patients will be referred to the TRANSFORM-HF Mortality Committee.

Table 2:

Death Event Triggers and Confirmations

	Confirmation Status
Source	Yes					No	Total
Trigger	Obituary	2nd Proxy	Med Rec	NDI	Other
Proxy	133	0	2	32	0	70	237
Online	81	0	2	1	0	2	86
Med Rec	NA	NA	0	0	NA	NA	0
Site	31	0	21	18	0	17	87
NDI	NA	NA	NA	19	NA	NA	19
Total	245	0	25	70	0	89	429

Obituary = obituary or grave marker; Online = online search; Med Rec = medical record Site = enrolling site

Figure 1 shows cumulative time to event results for 2019 deaths (n=150) jointly identified by the DCRI Call Center and the NDI. The Call Center Status and NDI Death Results show the cumulative death events identified by each method. Deaths only identified by the DCRI Call Center (n=12) or by NDI search (n=13) were excluded to avoid potential bias when comparing death ascertainment time measurements.

Figure 1: Time to Death Identification: Call Center vs. National Death Index.

Note: This figure is the Visual Take Home graphic

Median times to death event identification were 47 days (interquartile range, 11 to 103) for the call center and 270 days (interquartile range, 186 to 391 days) for the NDI.

DISCUSSION

Our results demonstrate a high level of agreement between DCRI Call Center and NDI identified patient death events in the TRANSFORM-HF trial. However, the time between a death event and its identification is significantly less for the call center. While the DCRI Call Center’s follow-up protocol has successfully identified follow-up events in the present and previous studies, these results may not be generalized to other call center approaches (6).

We believe both the DCRI Call Center and NDI are needed because they serve different functions. The DCRI Call Center identifies death events earlier and provides real time data (death and other) for use in monitoring study progress. The NDI searches both confirm call center triggered deaths and identify additional deaths not identified by the call center. If NDI searches were the sole death data source, this could significantly delay study results availability and the time when patients might benefit from their dissemination.

In 2017, only 29.8% of United States deaths occurred in hospitals; whereas, 8.3% occurred in hospice facilities, 20.8% in nursing homes, and 30.7% in patient homes (7). The location of death was similar for cardiovascular patients (8). Clearly, pragmatic clinical trials need death event ascertainment methods that are efficient and effective irrespective of location of death. While our results are encouraging, they will need to be validated in other settings. Because NDI data cannot be used for commercial purposes, it is important to rigorously validate the DCRI Call Center protocol and other death ascertainment methods that can be used by all study funders.

CONCLUSIONS

There is substantial agreement between deaths identified by a centralized call center and the NDI. However, the time between a death event and its identification is significantly less for the call center.

Highlights

• Death identification can cause unique challenges for pragmatic clinical trials that limit site follow-up activities to usual clinical care.

• This study evaluated the agreement between centralized call center death event identification and the United States National Death Index (NDI).

• There was substantial agreement between deaths identified by a centralized call center and the NDI (Cohen’s kappa, 0.920; 95%CI, 0.889 to 0.951).

• However, the time between a death event and its identification is significantly less for the centralized call center (median, 47 days) than for NDI (median, 270 days).

Lay Summary:

• Death identification can be challenging for pragmatic clinical trials that limit site follow-up activities to usual clinical care.

• This study evaluated the agreement between centralized call center death event identification and the United States National Death Index (NDI).

• There was substantial agreement between deaths ascertained by a centralized call center and the NDI. However, the time between a death event and its identification is significantly less for the centralized call center.