Dexmedetomidine Medication Records and Delirium After Coronary Artery Bypass Grafting

This is an updated analysis of a retrospectively registered observational cohort study. The clinical procedure eligibility period remains 1 January 2010 through 1 July 2025. Each live-network outcome batch is identified by its actual run date, saved query and analysis identifiers, native balance and outcome exports, data-return messages and available refresh metadata. Earlier analyses and technical refreshes are retained as distinct versions rather than treated as one immutable dataset.

The primary population uses a curated CABG definition comprising 10 CPT codes and 140 coronary ICD-10-PCS open-approach codes. The separate exploratory valve-coded population uses a curated 83-term CPT/ICD-10-PCS definition. Each population has its own procedure anchor, eligibility assessment and matching. Study enrollment encompasses the complete eligible populations before propensity-score matching, including patients who are not subsequently matched. A patient may qualify for both populations. No pooled effect estimate or formal interaction test is produced, and the valve analysis is not described as independent replication. Procedure codes do not establish clinically adjudicated isolated or first-ever surgery.

For the enrollment update, three aggregate queries were counted in one coordinated session on 9 September 2026, between 09:12 and 09:17 UTC, using the same network and five-domain healthcare-organisation filter. The complete CABG eligible definition returned 235,752 patients, the complete valve eligible definition returned 139,542, and an AND-combination retaining both complete definitions and their separate procedure-specific anchors returned an intersection of 32,182. The reported enrollment is the overlap-adjusted platform total: 235,752 + 139,542 - 32,182 = 343,112. This is a derived total, not an unadjusted sum, a matched sample size or a native cross-query union export. It assumes compatible platform counting units across the three sequential LIVE counts. No immutable snapshot identifier or query-level last-data-refresh timestamp was available, so the calculation does not establish an exact single-time-point union. Cross-HCO natural-person deduplication was not independently verified. The dated component counts, saved definitions and execution metadata are retained. Each outcome batch's before-matching, matched and analysed denominators are reported separately and do not redefine this enrollment total.

Age eligibility is applied to the qualifying procedure using Age at Event 18-100 years, subject to the platform's additional age-related privacy restrictions. Healthcare organisations must provide demographics, diagnoses, procedures, medications and laboratory domains. These restrictions apply to both exposure groups and both procedure populations; availability of five domains does not establish complete data for every patient.

Day 0 is the first qualifying procedure within the clinical period after the date and Age at Event filters, rather than the lifetime first procedure. The procedure group is selected as the analysis index in each exposure arm. Exposure is any RxNorm 48937 dexmedetomidine medication record on days -1 through 0. The comparator has no such record in that window; records outside the window, including later postoperative use, are permitted. Postoperative drug records do not move the index. Administration, dose, treatment indication and exact intraday sequencing are not established.

Patients with specified ESRD or dialysis records during days -90 through -1 are excluded. The platform displays 90 days as three fixed 30-day months. In both procedure populations, any R41.0 or F05 record on or before day 0 in all available prior history is excluded before matching, including records before 1 January 2010. All endpoints retain this eligibility specification. No broader secondary-outcome population is introduced, and continuous preoperative observation, complete 30-day follow-up or survival to a landmark is not required. Eligibility restrictions reaching day -1 or day 0 can occur after exposure has begun; applying them before matching does not remove the possibility of selection or collider bias. The analyses concern conditional record-based associations, not a total causal treatment effect.

Non-demographic propensity covariates use days -90 through -2, ending before the earliest exposure day. Age at index and sex are not assigned this window. The 29 constructs expand to 31 platform input rows: age at index; sex as male, female and unknown; 12 prespecified comorbidity indicators; six baseline medicine indicators; four dementia indicators (F01, F02, F03 and G30); opioid analgesics; benzodiazepine derivative sedatives/hypnotics; and categorical BMI, haematocrit and albumin. VA CN302 is not asserted to include all benzodiazepines. The operational BMI characteristic is TNX Curated 9083; the former 39156-5 identifier is not independently selectable in the current Characteristics interface, and equivalence is not assumed.

The saved categories are BMI <18.5, 18.5 to <25, 25 to <30 and ≥30 kg/m2; haematocrit <24, 24 to <30 and ≥30%; and albumin <3, 3 to <3.5 and ≥3.5 g/dL. Each lower boundary is inclusive and each upper boundary exclusive; the outer limits are unbounded. Albumin may be converted to g/L for presentation only. Platform display wording such as "At most" does not override these saved operators. The Characteristics heading referring to most recent observations is not treated as independent proof of the categorical matching membership rule. Category absence is not assumed to mean a normal result, complete measurement ascertainment or mean imputation.

The approved propensity-score specification uses native 1:1 logistic-regression propensity scores and greedy nearest-neighbour matching without replacement, with a caliper of 0.1 pooled standard deviations of propensity scores. TriNetX reruns matching automatically with each Outcomes Run. Consequently, each batch uses its own exported matched population, balance statistics and outcome denominators. Equal sample sizes or balance statistics do not prove identical patients or pairs across runs. Matching inputs and thresholds are not tuned according to outcomes. Each batch's native balance, retention, absolute standardised mean differences and denominators are reviewed before interpretation, without claiming these checks preceded that batch's automatic outcome calculation.

The primary outcome is a first qualifying R41.0 or F05 record during days 1-30 after the CABG-coded procedure. F05 alone is a sensitivity analysis retaining the same broader prior-delirium-free eligibility. Day 1 is the next calendar day, not a measured 24-hour interval or a verified alive-and-observable landmark. The primary test is the two-sided log-rank test. Kaplan-Meier estimates and complementary hazard ratios are interpreted with the platform's censoring and proportional-hazards diagnostics. Censoring after the last fact in the record does not independently establish every death or same-day tie rule. Where death is censored, one minus Kaplan-Meier survival is a net failure estimate, not competing-risk cumulative incidence. No competing-risk analysis is included in this revision. Exact interval-specific numbers at risk are unavailable through the native LIVE interface and are not reconstructed by multiplying survival probability by baseline cohort size.

All nine secondary endpoints use days 0-30. The eight inferential endpoints are all-cause mortality, N17 acute kidney injury, pneumonia, sepsis, a composite of blood transfusion, haemorrhage-related diagnosis or thoracic exploration codes, myocardial infarction, ischaemic stroke or transient ischaemic attack, and atrial fibrillation. Endpoint-specific prior-outcome exclusion is off, so nonfatal endpoints describe qualifying records rather than newly diagnosed disease in an independently restricted risk population. Same-day medication and outcome ordering is unknown. Non-proportionality is disclosed, with the hazard ratio interpreted as an overall summary rather than a constant effect.

The eight native two-sided log-rank tests form the fixed Benjamini-Hochberg family for the primary CABG secondary analyses. The primary outcome, F05-only sensitivity, valve exploratory analyses and descriptive endpoints are outside that family. Suppressed or unestimable tests are reported as unavailable without fabricating values or replacing the specified test. The family is not reduced according to results.

ICD-10-PCS 5A1955Z identifies a recorded ventilation code specifying more than 96 consecutive hours. It is described using patients with at least one record during days 0-30 and each batch's named denominator, not ventilation onset time or censoring-adjusted risk. No ventilation Kaplan-Meier or Cox analysis is performed. The separate days 1-30 encounter endpoint is the presence of HL7 v3 Visit Type IMP or CPT grouping 1013659. It does not require discharge followed by a new admission and is not readmission. For these two descriptive endpoints, unavoidable Measures of Association outputs are preserved and acknowledged as computed, but their comparative statistics are not treated as inferential study results or included in the multiplicity family. Record counts are not substituted for patient counts.

Coordinated refreshes are assessed using actual native run and refresh timestamps, not download filenames. A general notification that one or more HCOs did not return results is distinguished from an explicit incomplete-analysis error. Query-stage numbers of organisations queried and responding do not certify outcome completeness. Explicitly incomplete or unusable batches are retained as non-final evidence and are not silently replaced with older results. Technical recovery retains the approved scientific specification and is not selected according to effect direction or statistical significance.

Earlier analyses were known when the clinical definitions were revised. The automatic-rematching and descriptive-calculation amendments were approved after the first refreshed CABG primary/F05-only results had been viewed. Original registration, scientific decisions, technical amendments, saved runs, document approval and PRS release dates are recorded separately. No full pre-analysis lock or prospective registration is claimed retrospectively. Only permitted aggregate outputs and reproducibility materials are retained, and no retrospective observed-power claim is made.

Inclusion criteria

• A qualifying CABG-coded procedure for the primary analysis, or a qualifying valve-coded procedure for the separate exploratory analysis, during 1 January 2010 through 1 July 2025.
• Age at Event 18-100 at a qualifying procedure, subject to the platform's additional age-related privacy restrictions.
• A procedure-date index defined by the first qualifying procedure within the clinical period after the date and age filters, with the same eligibility logic in both exposure groups within each procedure population.
• Data contributed by healthcare organisations providing demographics, diagnoses, procedures, medications and laboratory domains.
• Medication-record status classified using RxNorm 48937 during days -1 through 0 relative to the qualifying index procedure.

Exclusion criteria

• An ESRD or dialysis record during days -90 through -1: ICD-10-CM N18.6, Z49 or Z99.2; SNOMED CT 265764009; ICD-10-PCS 5A1D60Z or 5A1D00Z; or ICD-9-CM 39.95.
• Any R41.0 or F05 record on or before day 0 in all available prior history relative to the index procedure, applied before matching in both the CABG-coded and valve-coded populations. Primary, secondary, F05-only and encounter endpoints retain this eligibility definition. Native matching is repeated within each outcome batch, and identical matched patients across batches are not established.

Some eligibility records may occur after the exposure window begins on day -1, which may introduce selection or collider bias despite application before matching. The study does not require continuous preoperative observation, complete 30-day follow-up or survival to a landmark. Missing laboratory measurements do not automatically exclude a patient. The available aggregate outputs do not permit clinical adjudication of individual procedures or confirmation that surgery is first-ever or isolated.