HardWindow Functions High Frequency

Patient ICU Heart Rate Spike Detection with LAG()

Frequently Asked By:AppleGoogleAmazon

Interactive In-Browser Workspace

Practice this problem on live schema tables in PostgreSQL 16 WASM with instant row verification and visual EXPLAIN plans.

Solve in Workbench

Problem Statement

For each vital reading in `icu_telemetry`, compute the heart rate change compared to the patient's immediately preceding reading. Calculate: 1. `patient_id` 2. `reading_time` 3. `heart_rate` 4. `prev_heart_rate`: Heart rate from the preceding reading using `LAG()` partitioned by `patient_id` 5. `bpm_delta`: (`heart_rate - LAG(heart_rate) OVER (PARTITION BY patient_id ORDER BY reading_time)`) Order results by `patient_id` ascending, then by `reading_time` ascending.

Why Technical Interviewers Ask This

ICU monitoring systems continuously evaluate bedside telemetry. Sudden increases of 20+ beats per minute (BPM) between consecutive readings trigger cardiac telemetry alarms.

Common Pitfalls & Interview Traps

Forgetting required sorting and deterministic tie-breaking.

Not accounting for NULL or zero-value denominators during calculations.

Using inefficient quadratic subqueries instead of single-pass window functions constructs.

Canonical Modern SQL Solution

PostgreSQL 16 / ANSI SQL

SELECT 
        patient_id,
        reading_time,
        heart_rate,
        LAG(heart_rate) OVER (PARTITION BY patient_id ORDER BY reading_time) AS prev_heart_rate,
        (heart_rate - LAG(heart_rate) OVER (PARTITION BY patient_id ORDER BY reading_time)) AS bpm_delta
      FROM icu_telemetry
      ORDER BY patient_id ASC, reading_time ASC;

Approach Breakdown

This query addresses the problem requirements by filtering and structuring data in Window Functions. It meets the target criteria: Patient telemetry readings showing previous heart rate and BPM delta.

Edge Cases to Verbally Clarify with Your Interviewer

The dataset contains zero matching records (must return valid empty result without crashing).

Multiple records tie on sorting criteria (must evaluate tie-breaking order deterministically).

Foreign keys or secondary attributes have NULL values.

Frequently Asked Follow-Up Questions

What makes this question a favorite among Apple and Google interviewers?

Interviewers use this problem to test whether candidates write defensive, scalable Window Functions queries that run efficiently on multi-million row production databases without excessive memory usage.

How does the query optimizer execute this query?

Using single-pass table scans with index lookups or sort buffers. Running EXPLAIN (ANALYZE, BUFFERS) in the DB Exercises workbench visualizes the execution nodes and buffer hit cache.