Data sources, data elements and linkage

Table 1 shows study data elements and sources and time points of data collection.

Data elements and sources for key trial outcomes by study procedure

ACPE, Advance Care Planning Engagement Questionnaire; CMS, Centers for Medicare and Medicaid Services; DCS, Decisional Conflict Scale; eGFR, estimated glomerular filtration rate; EHR, Electronic Health Record; EuroQol, 5-level EuroQol-5D version (EQ-5Dimension-5Level); KD-QoL-36, Kidney Disease Quality of Life; PN, primary nephrologist; RA, research assistant; SPMSQ, Short Portable Mental Status Questionnaire; SQ, Surprise Question; YDDAU, Yorkshire Dialysis Decision Aid Usefulness Scale.

Data on sociodemographics including age, gender, race, ethnicity, primary language, health insurance, education, marital status, religion and religious attendance will be assessed via surveys.

Will include any documentation in the EHR reflecting an ACP conversation (completion of advance directive or Physician’s Orders for Life Sustaining Treatment (POLST); code status documentation; provider note reflecting ACP discussion) (primary outcome). The primary analysis will be based on the EHR notes of the nephrology clinic team. In the secondary analysis, we will also add notes from other providers.

We will ask, via RA administered survey, four validated questions regarding ACP engagement.⁷⁵ (How ready are you to talk to your caregiver? To your doctor? To appoint a surrogate? To sign an ACP document?)

Resuscitation preferences regarding CPR (yes, no or unsure) and dialytic versus non-dialytic treatment (haemodialysis, peritoneal dialysis, medical management without dialysis or unsure) will be assessed after randomisation, and every 2 months until the end of study follow-up at 12 months or death.

We will survey patients regarding whether they have had prior ACP discussions.

We will also measure disease-specific quality of life using data obtained from the Kidney Disease Quality of Life (KDQOL-36)⁷⁶ administered at baseline and every 60 days thereafter. Responses to each of the 36 items will be scored (0–100) and the overall mean used as the quality of life measure for each survey round.

To capture differences in quality of life, besides longevity, associated with the choice of kidney care approach, we will use EuroQol’s EQ-5D-5L instrument as the quality measure.⁷⁷ This instrument takes responses to five questions on mobility, self-care, ability to perform usual activities, pain and anxiety/depression to produce a validated quality score (0–1). This instrument will be administered at baseline and every 60 days thereafter. The cumulative quality scores from consecutive rounds of survey will be used to obtain the QALYs for the exposure time.

We will measure decisional conflict using the Decisional Conflict Scale (DCS), which attempts to measure decisional uncertainty.⁷⁸

For those patients randomised to the video intervention, we will measure, via survey, acceptability of the decision aid using a modified version of the validated Yorkshire Dialysis Decision Aid Usefulness Scale.⁷⁹ We will also ask questions regarding comfort viewing the video, which we have validated in our prior work.^{38 47 48 50 52 54–56 80}

All patients will be asked their preferences for kidney failure care at baseline. We will then assess their follow-up preferences by chart review in the electronic medical record.

The main source of differences in costs between the video and control arms will be from the stream of healthcare services used, including that for kidney failure care, over the exposure time. Based on prior evidence of healthcare spending of CKD patients, we will identify the major components of services used, including inpatient, pharmacy, outpatient, emergency department and dialysis.^{21 81} We will also examine utilisation by subgroups with comorbidity of diabetes, heart failure and cardiovascular disease. We will use Medicare claims data to obtain the associated costs, including payments by Medicare and secondary payers (eg, out-of-pocket payments).⁸² Medicare claims data are available for a majority of Medicare enrollees (about 75% choose the Fee for Service plan). As these data are unavailable for the others who choose managed care plans or are enrolled from the Veterans Affairs, we will impute the costs (per year) based on the average costs for the Fee for Service participants separately by the type of kidney care chosen.⁸¹

We will conduct NLP-assisted EHR review for documentation of ACP (primary outcome). This EHR review will include keyword-based searches for documentation of limitations to life-sustaining treatment, goals of care, healthcare proxy designation or communication on the patients’ behalf, palliative care involvement, hospice preference or utilisation, discussions surrounding dialytic versus non-dialytic therapies (including time-limited trials of dialysis), as well as completion of any advance directive and/or POLST. For patients who die prior to 12 months, we will conduct an NLP-assisted EHR review to assess ACP documentation (primary outcome), type of kidney failure treatment received prior to death, receipt of palliative care, hospice, or CPR/intubation in the last month of life, and place of death (eg, intensive care unit, home, etc).

NLP-assisted EHR review will rely on the ClinicalRegex software, which allows for rapid semi-automated clinical note review. ClinicalRegex presents operators with clinical notes highlighted in particular areas located by keywords associated with the concepts in question. Site operators will then ensure that keywords found within the notes appear in the correct clinical context (as in the documentation of ACP conversations). This method will be used at each site to search all collected outpatient clinical documentation data from the EHR for ACP documentation, similar to prior studies using NLP.^83–85

For each NLP domain (ie, goals-of-care discussion, limitations to life-sustaining treatment), we have built a keyword library with the goal of identifying relevant documentation within clinical notes. Each keyword library will be refined and validated by the review of retrospective clinical notes in each site’s local EHRs to generate formal metrics (accuracy, sensitivity, specificity, etc) across all sites.⁸⁶

All site operators who will be engaged in this NLP must participate in training on note annotation practices and must demonstrate proficiency in annotating notes containing clinical concepts expected to be found during this trial. Proficiency will be determined by the use of a calibration test consisting of 20 mock clinical narratives which will be used to cross-validate annotation practices across all sites.

The EHR data will be reviewed by Dana-Farber Cancer Institute (DFCI) data staff and unblinded investigators. The NLP results and metadata (keyword frequencies, rates of agreement between annotator and keyword library) for each domain will be used across all sites to identify out of range or unexpected results, and a summary will be sent to each site. Conference calls will be conducted with relevant investigators and programmers to adjudicate any issues. We will then finalise NLP analysis results and submit to the study statistician for further analysis.

We have data use agreements from all sites to ensure adherence to the process and procedures for the protection of human subjects and protected health information (PHI). We will collect the minimum PHI needed from study participants and store all study information on HIPAA-compliant, password secured servers. We will separate participant identifying information from password-secured files while maintaining a linkage file at study sites. The linkage file will be restricted per local rules for PHI. We will transfer study data through HIPAA-secure methods specific to each site. Data will be sent to DFCI for data management and to Boston Medical Center and Massachusetts General Hospital for analysis. The final data set will be available to trial investigators on completion of the study and others can be provided access on reasonable request.