Interview guides · HiLabs

HiLabs Data Scientist Interview Questions & Guide 2026

Timed data scientist practice for HiLabs: short SQL and probability sets run end to end, then scored against a rubric.

The 60-second brief

As a Data Scientist at HiLabs, you sit at the intersection of complex healthcare data and cutting-edge artificial intelligence. Your work is fundamental to the company’s mission: transforming messy, real-world healthcare data into actionable insights that improve patient outcomes and operational efficiency at scale. You are not just building models in a sandbox; you are responsible for deploying production-grade systems that handle massive datasets, requiring a balance of rigorous statistical thinking and robust engineering discipline.

Product-sense cases reward reasoning from a mechanism to a testable prediction. Reciting every metric you can name reads as pattern matching; naming the single quantity that would move if your explanation were true reads as thinking.

Focus areas

Reconstruct drug exposure intervals from dispensing records
Build member-month denominators from overlapping coverage spans
Correct for claims runout before reporting recent months

Overview

This role offers the opportunity to tackle some of the most difficult challenges in the healthcare domain, from predictive modeling to large-scale data transformation. You will collaborate closely with product and engineering teams to ensure that your solutions drive measurable business impact. Success in this role requires a high degree of ownership, technical versatility, and the ability to articulate complex analytical concepts to non-technical stakeholders. You will be expected to thrive in a fast-paced environment where your output directly influences the trajectory of HiLabs products.

The HiLabs interview process

HiLabs candidates report 5 rounds · ≈ 4-6 weeks. The stages below are what candidates describe, not a published process.

Technical Screening

Much of what gets scored here happens out loud while you type. Nobody can see your reasoning inside a half-written query, so five silent minutes read as being stuck even when they are not. State the plan in plain language first: which tables, what grain you are aggregating to, and the one filter that defines the population. Then write it. The narration doubles as insurance, because a wrong plan gets caught early and cheaply while a wrong query gets caught at the end with no time left to redo it. A timed statistics section, where one exists, is a separate test with its own clock.

What to demonstrate

Whether the query you write matches the plan you just described
What you do with a hint, meaning whether the correction gets absorbed or the first approach gets defended
Whether you can debug your own wrong output by reading the result set and naming which part of the query produced the anomaly

How to prepare

Solve three problems while screen-sharing into a recording, then watch it back and mark every stretch longer than thirty seconds where you said nothing
Practise compressing the plan into one sentence before typing, then check afterwards whether the finished query actually matched it
Time yourself on statistics questions that carry a business reading, such as what a confidence interval does and does not claim, rather than re-reading notes without a clock

Machine Learning Deep-Dive

An extra round usually exists because something is still open after the standard loop: a skill the earlier interviews did not sample, a level decision, or two interviewers who disagreed. It is rarely a rerun of what you already did well. Ask the recruiter who you are meeting, what function they sit in, and how long the session runs. That is an ordinary scheduling question, and the answer changes what you should prepare. What separates a strong candidate here is treating the round as a fresh evaluation with its own bar, rather than assuming earlier performance carries you through or sinks you.

What to demonstrate

Whether you can answer well on ground the earlier rounds did not cover, without leaning on what you already said to someone else
Consistency of the facts in your stories: the same sample size, timeframe, team size and scope of your own role as in earlier conversations
How you handle an unfamiliar format live, including whether you ask what kind of answer is wanted before producing one

How to prepare

Ask the recruiter for the interviewer's function, the length, and whether to expect a coding surface, a discussion, or a presentation. Preparing for a 30 minute conversation with a partner team is not the same work as preparing for a 60 minute technical block.
Write out what each earlier round actually covered, then list the two or three areas nobody probed. That gap is the most likely subject of the extra round.
Re-read the numbers in the project stories you have already told, so a second telling does not quietly contradict the first.

Statistics Assessment

This round decides whether someone can hand you a schema and a question and trust the number that comes back. Correctness under a clock is the bar, not clever syntax. The habit that separates strong from weak answers is checking the grain: after every join, know how many rows you expect and whether the count moved. Most wrong answers in this format are not wrong logic, they are a fan-out from a key that turned out not to be unique, or a filter applied before an aggregate when it belonged after. Say what you expect before you run it.

What to demonstrate

Whether your row counts survive each join, and whether you notice on your own when they do not
Deliberate handling of rows that fail to match, including whether the question needs an inner join or a left join with the non-matches kept and counted
Whether NULLs are treated on purpose, given that a NULL compares equal to nothing and that COUNT of a column skips it
Reaching a defensible answer inside the window instead of a refined one after it

How to prepare

Take a two-table schema, write a join that fans out on purpose, then fix it by collapsing the many-side to one row per key before joining. Repeat until the fix is reflex rather than recall.
Write a funnel as one query and print the distinct user count at each stage, then confirm each stage is a subset of the one above it rather than assuming it
Do a few timed runs in a plain text box with no autocomplete and no formatter, since assessment editors often have neither

Cultural Fit Interview

An added round often puts you in front of someone outside the core hiring team: a partner engineer, a product owner, a domain expert, sometimes a more senior manager. The question they are really asking is not whether you can do the work but whether they would trust a number that came from you. That changes what a good answer looks like. Lead with what the decision cost and what it changed, keep the method available but not central, and be plain about the limits of your evidence. Overstating a result is the fastest way to lose this round.

What to demonstrate

Whether you can explain a technical choice to someone who will never read your code, without either flattening it into nothing or hiding inside jargon
Honesty about evidence strength: what the analysis establishes, what it only suggests, and what it cannot say at all
How you take disagreement, specifically whether you update on a good objection, hold your position with reasons, or fold on contact

How to prepare

Write the two-sentence version of your most technical project for a non-specialist, then check that neither sentence needs a method name to make sense.
For one result you are proud of, write the strongest objection someone could raise and a response that concedes the part of it that is correct.
Prepare one decision that turned out to be wrong: how you found out, what it cost, and what you changed afterwards. A senior cross-functional interviewer asks for this more often than a technical one does.

Project Discussion

Rounds outside the standard loop often open with something deliberately under-specified: a loose business problem, an open question about a product area, a dataset described in one sentence. The common failure is surveying, listing six plausible approaches and committing to none of them. The thing that separates a strong answer is scoping out loud. State what you are treating as the goal, name the metric you would move, say what you are choosing not to do and why, then take one path through to an actual answer. An interviewer can follow you down a narrow path. Nobody can grade a menu.

What to demonstrate

Whether you turn an ambiguous prompt into a stated question with a measurable outcome before doing any work
The judgement visible in what you cut, and whether you say why you cut it rather than silently dropping it
Whether you land on a concrete recommendation with its caveat attached, rather than an unranked set of options

How to prepare

Take three vague prompts, such as 'is this feature working', 'why did retention drop', and 'should we expand into a new segment'. For each, write one sentence of goal, one primary metric with its window, and two things you are explicitly not doing.
Practise giving the recommendation first and the reasoning second, in five minutes. Loosely defined rounds are usually time-boxed, and an answer that arrives last often does not arrive.
Keep a running assumption list as you talk, on paper or in the shared doc, so the interviewer can challenge one assumption instead of your whole answer.

Where candidates lose points

Rates built on member counts rather than exposure

Members join and leave mid-period, so dividing events by distinct members mixes a person covered for 30 days with one covered for 365. New joiners also have artificially low observed utilisation because their claims have not arrived yet and because care takes time to initiate. Denominators must be member-months or member-years, and comparative quality measures usually need a continuous-enrolment requirement with an explicit allowable gap, stated in days.

Treating clinical measurements as missing at random

A lab result, a vital sign, or a screening exists because someone ordered it, and ordering tracks suspicion of disease, visit frequency, and site workflow. Imputing the mean or dropping incomplete rows biases the population estimate and can flip the sign of an association, because the untested are systematically healthier or systematically disengaged. The presence indicator is often more predictive than the value, which is a warning sign rather than a feature win: a model that learns test ordering will not transfer to a site with different protocols.

Reporting a mean for a heavy-tailed metric without saying what it hides

For spend, session length or items per order, a small fraction of units carries most of the total, so the mean has a wide standard error and one account can move it. Fix the handling before you see the result: cap or winsorise at a pre-declared percentile, and report the median or the share above a threshold next to the mean. Capping changes the estimand, so say which question the capped number answers, and check how much of any difference comes from the top 0.1 percent of units.

Reading experiment results before checking the arm split

Compare observed arm counts against the intended allocation ratio, not an assumed even split, and set the alarm far below the conventional 0.05: at 0.05 roughly one healthy experiment in twenty trips it, which is why sample-ratio checks usually run at p < 0.001 or stricter. The test's power scales with sample size, so it misses a real diversion on a small experiment and fires on an imbalance too small to move the estimate on a very large one. A flag means go find the assignment or logging fault before reading any outcome, not report a mismatch.

Questions & practice

Explain the difference between correlation and causation in the context of user behavior on our platform.

Medium · statistics and probability

Approach

Quantify uncertainty explicitly rather than reporting a point estimate alone.
Translate the result into the decision it informs, in one plain sentence.
Sanity-check the answer against a simple bound or a simulated case.

Follow-up

What sample size would you need to detect an effect half this size?
Which assumption here is most likely to be violated in practice?

Proportion of days covered with shifted, truncated refill intervals

Hard · interval merging · adherence · path dependence

pharmacy_claim has fill_id, member_id, therapeutic_class_code, fill_date, days_supply, reversal_flag, reversed_fill_id. For one therapeutic class and a fixed 12-month window, compute proportion of days covered per member: distinct days on which a dispensed days_supply covers the day, divided by days from the member's first in-window fill through the window end. An early refill shifts coverage forward rather than stacking, and coverage is truncated at the window end. Drop both rows of every reversed pair. Return member_id, pdc, and the share at pdc >= 0.80 among members with at least 2 fills and at least 91 days of follow-up.

Approach

Remove reversals as pairs first. Drop every row with reversal_flag true, and also drop the fill_ids those rows point at through reversed_fill_id. Dropping only the flagged row leaves a dispense that was never collected in the exposure.
Walk fills per member in fill_date order carrying a cursor: start = max(fill_date, previous_end + 1 day), end = start + days_supply - 1. The shift is path dependent, so a plain cumsum over days_supply does not reproduce it. Use itertools.accumulate or a per-member loop over numpy arrays, not a row-wise apply over the whole frame.
Truncate the last interval at the window end before measuring. Without truncation a 90-day fill dispensed on the final day pushes the covered-day count past the denominator and PDC above 1.0.
Use the denominator the definition states: first in-window fill_date through window end, inclusive. Not a flat 365, and not first fill to last fill, which is a different metric that rewards early discontinuation.
Apply the eligibility filter before computing the >= 0.80 share, and report the size of that denominator next to the share. A share without its denominator is not reviewable.

Follow-up

How does PDC differ from medication possession ratio, and which of the two can exceed 1.0?
A member switches to a different ingredient inside the same therapeutic class mid-window. Should the intervals chain, and what does that do to the class-level number?
Members who die or lose coverage mid-window get a short denominator and often a high PDC. If you then compare mortality by adherence category, what bias have you built in and how do you remove it?

Simulate how ignoring clustering inflates the false positive rate

Medium · simulation · design effect · type i error

Simulate a provider-level comparison with no true effect. 40 providers, 60 members each, a continuous outcome with total variance 1 and intraclass correlation 0.02. Randomise providers 20 to 20, not members. Over 5,000 replications, report the share of replications where a two-sample t-test run on all 2,400 individual observations, ignoring provider, returns p below 0.05. Report the same share when the unit of analysis is the 40 provider means. Then state how the first number follows from the design effect 1 + (m - 1) x ICC.

Approach

Decompose the variance explicitly rather than tuning it: provider variance = ICC = 0.02, residual variance = 1 - ICC = 0.98. Draw u_j once per provider and e_ij per member, so the outcome is u_j + e_ij and the correlation between two members of one provider is 0.02 by construction.
Randomise at the provider level. That is the entire mechanism: the treatment indicator is constant within a provider, so in any single replicate the provider intercepts are confounded with the arm, and the naive test reads that confounding as signal.
Vectorise over replications with a (reps, providers, members) array. 5,000 x 2,400 normals is trivial in numpy, and a Python loop is what tempts people to cut replications to 500 and report a number with Monte Carlo noise of plus or minus 0.017.
Predict the answer before running it: design effect = 1 + (60 - 1) x 0.02 = 2.18, naive standard errors are too small by sqrt(2.18) = 1.476, so the rejection rate at nominal 0.05 is about 2 x (1 - Phi(1.96 / 1.476)) = 0.18. Agreement between prediction and simulation is what makes the result a demonstration rather than an anecdote.
Run the provider-mean analysis as the control. Recovering 0.05 there proves the generator is correct and isolates the failure to the analysis unit.

Worked solution (30 min)

Draw u with shape (reps, 40) from N(0, sqrt(0.02)) and e with shape (reps, 40, 60) from N(0, sqrt(0.98)); y = u[:, :, None] + e.
Assign providers 0-19 to one arm and 20-39 to the other; because there is no true effect, a fixed split is valid and removes one source of Monte Carlo noise.
Naive test: flatten each arm to 1,200 observations and compute a two-sample t statistic per replication with vectorised means and pooled variance.
Cluster test: average within provider to 40 values, then run the same two-sample t on 20 versus 20.
Report both rejection shares and compare the naive one against 2 x (1 - Phi(1.96 / sqrt(2.18))).

Expected result: The naive individual-level test rejects in roughly 18 percent of replications, against a nominal 5 percent. Monte Carlo standard error at 5,000 replications is about 0.005, so anything in 0.17 to 0.19 is consistent. The provider-mean analysis rejects in about 5 percent, within roughly 0.003.

Verify it yourself

Set ICC to 0 and both analyses return about 0.05. If the naive one does not, the data generator is wrong, not the test.
Raise members per provider from 60 to 120 with ICC fixed and the naive rate rises, because the design effect grows linearly in cluster size.
The provider-mean analysis stays near 0.05 at every ICC you try, which is the definition of a valid test.
The simulated naive rate lands within about 0.01 of 2 x (1 - Phi(1.96 / sqrt(design effect))).

Follow-up

How many providers would you need for 80 percent power on a 0.2 SD difference with 60 members each and this ICC?
Cluster sizes are unequal in reality. The design effect approximation becomes 1 + ((1 + CV^2) x mbar - 1) x ICC. Which direction does that move your sample size and why?
The intervention cannot be withheld from any provider. Sketch a design that still yields a defensible estimate.

Describe how you would optimize a slow-running SQL query that joins several million-row tables.

Medium · sql and data manipulation

Approach

Check whether any join is one-to-many before aggregating, or the sums inflate.
Compute rates by summing numerator and denominator separately, never by averaging rates.
Handle the rows that do not match: a LEFT JOIN with a NULL check is usually the question.

Follow-up

How would you verify this result without re-running the same query?
How does the query change if the join becomes one-to-many?

Write a query using SQL window functions to identify the top three healthcare providers by claim volume per region.

Medium · sql and data manipulation

Approach

Check whether any join is one-to-many before aggregating, or the sums inflate.
Say which table is the grain you start from, and join outward from it.
Handle the rows that do not match: a LEFT JOIN with a NULL check is usually the question.

Follow-up

How would you verify this result without re-running the same query?
What breaks if events arrive late or out of order?

How would you handle missing values or inconsistencies in a large, messy medical claims dataset?

Medium · sql and data manipulation

Approach

Say which table is the grain you start from, and join outward from it.
Compute rates by summing numerator and denominator separately, never by averaging rates.
State the window function and its partition and ordering out loud before writing it.

Follow-up

How does the query change if the join becomes one-to-many?
What breaks if events arrive late or out of order?

Report site activation to first randomization without hiding non-enrollers

Medium · grain collapse · left join · censoring

study_site holds one row per (study_id, site_id): site_activation_date, which is NULL for a site selected but never activated, and site_status. study_subject_visit holds subject_visit_id, study_id, site_id, subject_id, randomization_ts (NULL for screen failures), visit_name, visit_number, actual_visit_date and visit_status, with one row per subject per protocol visit, so randomization_ts repeats across every visit of a randomized subject and a site that has screened nobody has no rows there at all. For one study_id and a stated data cut date, return per activated site the activation date, the first randomization_ts and the days between them, with activated sites that have randomized nobody still present. Add a study-level summary giving median days to first randomization and the share of activated sites with no randomization by day 90.

Approach

Take the site universe from study_site, never from study_subject_visit. The visit table can only contain sites that have screened at least one subject, so a universe built from DISTINCT site_id there omits exactly the sites this metric exists to expose. Filter study_site to the study and to site_activation_date IS NOT NULL, then LEFT JOIN the per-site first randomization.
Collapse the grain before aggregating anything on the visit side. The table repeats randomization_ts across every visit of a randomized subject, so any count taken at row grain is inflated by visits per subject. MIN(randomization_ts) per site is safe, COUNT is not.
State the date boundary rule. randomization_ts is a timestamp and site_activation_date is a date, so a randomization at 23:40 local time can land on either side of a day boundary depending on the zone the timestamp is stored in. Cast once, deliberately, and say which zone.
Use PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY days) for the median and be explicit that it ignores NULLs, so it is a median over enrolling sites only. That is exactly why the non-enroller share is reported beside it rather than instead of it.
Censor the day-90 denominator against the data cut. A site activated 40 days before the cut has not had 90 days to randomize, so it cannot be scored on that bar and belongs outside that denominator.

Worked solution (25 min)

CTE sites: site_id and site_activation_date from study_site for the study where site_activation_date IS NOT NULL. The (study_id, site_id) key already gives one activation date per site, so assert that key rather than de-duplicating defensively.
CTE first_rand: site_id and MIN(randomization_ts) over study_subject_visit rows for the study where randomization_ts IS NOT NULL, grouped by site_id.
LEFT JOIN sites to first_rand and compute days_to_first_randomization with an explicit cast of the timestamp to a date.
Summary: PERCENTILE_CONT(0.5) over non-null days, plus the count and share of sites with no randomization by day 90, restricted to sites activated at least 90 days before the data cut.
Compare the site-level row count with COUNT(DISTINCT site_id) in study_subject_visit for the study; the difference is the activated sites that have screened nobody, which is the population the metric is for.

Expected result: One row per activated site with site_id, site_activation_date, first_randomization_ts and days_to_first_randomization, NULL in both randomization columns for sites that have randomized nobody, including sites with no study_subject_visit rows at all; plus one summary row with median_days_over_enrolling_sites, sites_activated_90d_or_more and share_with_no_randomization_by_day_90.

Verify it yourself

Site-level row count equals the number of study_site rows for the study with a non-null site_activation_date, and is greater than or equal to COUNT(DISTINCT site_id) in study_subject_visit for that study. Below it means either an activated site is missing from study_site or a visit row carries a site_id whose activation date is NULL, both data-quality findings.
No days_to_first_randomization is negative. A negative value is a data-quality finding to raise, not a number to clamp at zero.
The day-90 denominator excludes sites activated within 90 days of the data cut. Including them can only add sites that have not had a full 90 days, and any of those without a randomization would be miscounted as a failure.

Follow-up

The same site_id appears under two studies. What breaks, and what is the correct grain for this metric?
Would you present this as a median, as a Kaplan-Meier curve of time to first randomization, or both, and what does the curve show that the median cannot?
A site randomized one subject who withdrew at visit 2. Does that site count as enrolling here, and how does that differ from a randomized-and-retained measure?

If we notice a sudden 10% drop in user engagement on our main dashboard, how would you go about diagnosing the root cause?

Medium · product sense and metrics

Approach

Name one primary metric, then the guardrail that stops it being gamed.
State what result would change your recommendation, so the answer is falsifiable.
Fix the population and the time window before naming any metric.

Follow-up

How would you detect that the metric is being gamed rather than genuinely improving?
Which segment would you cut first, and what would that rule out?

How would you define "data quality" for a healthcare dataset, and what metrics would you use to track it?

Medium · product sense and metrics

Approach

Restate the decision this analysis has to support, and who acts on the answer.
Fix the population and the time window before naming any metric.
Decompose the metric into the rates that drive it, and say which one you would check first.

Follow-up

Which segment would you cut first, and what would that rule out?
What would you do if the primary metric and the guardrail moved in opposite directions?

How would you design a metric to measure the success of a new data validation feature in our healthcare platform?

Medium · product sense and metrics

Approach

State what result would change your recommendation, so the answer is falsifiable.
Decompose the metric into the rates that drive it, and say which one you would check first.
Fix the population and the time window before naming any metric.

Follow-up

Which segment would you cut first, and what would that rule out?
How would you detect that the metric is being gamed rather than genuinely improving?

What are the common experimentation pitfalls you have encountered when running A/B tests in a production environment?

Medium · a/b testing and experimentation

Approach

Decide the analysis before seeing data, including how long it runs and when you look.
Say whether units interfere with each other, and switch design if they do.
Name the randomisation unit first; it decides the variance and what the test can detect.

Follow-up

What would you conclude if the result is positive but the test is underpowered?
What would you do if you could not randomise at all?

Explain how you would address the issue of "novelty effect" in a product experiment.

Medium · a/b testing and experimentation

Approach

State the primary metric and the minimum effect worth shipping, then size the test.
Say whether units interfere with each other, and switch design if they do.
Name the guardrails that would stop a launch even on a positive primary result.

Follow-up

How would you handle interference between treated and control units?
What would you do if you could not randomise at all?

Diagnose a sample ratio mismatch in a randomised outreach test

Easy · sample ratio mismatch · randomisation audit · selection bias

A member-level outreach test hashed member_id into two arms and wrote assignments to a bucket table before any contact was made. The analysis extract, built by joining assignments to member_enrollment and to the vendor send file, contains 20,560 treatment and 19,440 control members out of 40,000 assigned. The intended split was 50/50. State the test you would run on the split, compute it, say what it means for the outcome analysis, and describe how you would locate the cause.

Approach

Test the split against the intended allocation with a chi-square goodness-of-fit on one degree of freedom, or equivalently a two-sided normal test on the treatment count with standard deviation sqrt(N x 0.5 x 0.5). Both give the same p-value.
Stop the outcome analysis before running it. A significant ratio mismatch means the analysis population is not the randomised population, so any treatment effect is confounded by whatever selected members out, and no covariate adjustment can be trusted to fix it.
Instrument the pipeline stage by stage and recompute the ratio at each one: the assignment table, the join to member_enrollment, the vendor send file, the response join, and the final extract. The mismatch appears at exactly one stage, and that stage is the bug.
Given this domain, check the eligibility join first. Coverage terminations are back-dated and land after the fact, so a join that requires active coverage at extract time silently drops members, and the drop is not independent of arm if the outreach itself triggered contact that refreshed eligibility records.
Compare the dropped members with the retained ones on termination_date, termination_reason_code, product_type, and prospective_risk_score to characterise the selection, then rerun the test on a corrected extract rather than rebalancing the arms.

Worked solution (15 min)

Expected count per arm is 20,000.
z = (20,560 - 20,000) / sqrt(40,000 x 0.5 x 0.5) = 560 / 100 = 5.6.
Equivalently chi-square = 560^2 / 20,000 + 560^2 / 20,000 = 15.68 + 15.68 = 31.36 on 1 df.
Convert to a p-value: two-sided p is about 2e-8.
Pull arm counts at each pipeline stage and find the first stage where the ratio departs from 50/50.

Expected result: z = 5.6, chi-square = 31.4 on 1 degree of freedom, p is about 2e-8. This is not chance. The experiment is not analysable until the differential drop is located and either fixed or demonstrated to be independent of assignment.

Verify it yourself

A deterministic hash on member_id should produce an exact or near-exact 50/50 split at the assignment table; if it does not, the randomiser itself is the bug.
The imbalance should appear at exactly one pipeline stage, not drift gradually across several.
The dropped members should differ from the retained on at least one observable, which both confirms the selection and describes it.

Follow-up

The assignment table itself is exactly 20,000 / 20,000. What does that tell you and where do you look next?
A colleague proposes randomly downsampling the treatment arm to 19,440 to restore balance. Why is that not a fix?
At what imbalance would you not investigate, and how did you pick that threshold?

Readmission rate rose while every facility improved

Medium · simpsons paradox · mix shift · risk adjustment

Quarter over quarter the 30-day unplanned readmission rate for an attributed population rose from 14.2 to 15.1 percent, yet all eleven facilities in encounter show a flat or falling rate over the same period. You have encounter (facility_id, admit_ts, discharge_ts, encounter_type, admission_type, discharge_disposition, drg_code, is_planned_admission, index_encounter_id) and twelve months of prior medical_claim_line for comorbidities. Explain the direction reversal numerically, and state what the population-level number should be reported as instead.

Approach

Confirm the reversal is arithmetic, not a definition difference, by recomputing both the facility rates and the pooled rate from the same query with the same index-stay exclusions.
Decompose the pooled change exactly with a Kitagawa split. With facility weights w and within-facility rates r, the change equals the sum over facilities of (w_1 - w_0) times (r_0 + r_1)/2, the mix term, plus the sum of (w_0 + w_1)/2 times (r_1 - r_0), the within term. If the mix term exceeds the total change and the within term is negative, the reversal is volume shifting toward facilities with higher baseline rates.
Ask why the weights moved. Check index-stay volume by facility, and check whether an exclusion changed behaviour: acute-to-acute transfers (discharge_disposition 'transfer_acute') must not open an index stay, and if one facility began transferring more, its denominator shrinks toward its sickest retained patients.
Check case mix inside the facilities too, using drg_code and comorbidities drawn from the 12 months of claims before admit_ts, excluding complications of the index stay itself. A facility rate can be flat while its case mix worsens.
Replace the raw pooled rate with an observed over expected ratio, where expected is the sum of predicted probabilities from a model fitted on age, sex and prior comorbidities. Raw rates are not comparable across a population whose facility and case mix is moving.

Follow-up

Your O/E ratio is 1.04 with a 95 percent interval of 0.96 to 1.13. What does that let you say to a facility, and what does it not?
If one facility contributes 40 percent of index stays, how does that change how you present the decomposition?
How would you handle a member with two index stays in the quarter, where the second is also a readmission from the first?

Suggested practice checklist

Four days spend equal time on query work, statistics, modelling and product judgement at deliberately shallow depth, which produces a scored map of where you actually stand. The last three days spend everything on the two areas the role weights most, and close by re-running day one to measure movement.

Day 1 — Breadth pass: query fluency

Solve six prompts spanning aggregation, joins, window functions and date arithmetic in 60 minutes total, stopping at 10 minutes each whether or not it works, and mark every prompt as solved, solved slowly, or stuck.
For each unsolved prompt write the single blocking sentence (I lost the grain, I did not know the frame clause, I could not express the date boundary) instead of reading the solution.
Translate one pandas transformation you know well into SQL and one SQL query into pandas, checking that both return the same row count and the same totals.

Deliverable: A scored six-row table, one line per prompt, saved for the day-seven re-run.

Day 2 — Breadth pass: statistics and inference

Answer ten short questions in writing with nothing open: what a p-value is conditional on, what a 95 percent interval covers across repeated samples, when a paired test is the right one, what the bootstrap estimates, why multiple comparisons inflate false positives, how controlling the family-wise error rate differs from controlling the false discovery rate, what power depends on, what a missed real effect costs a product, the three situations where the central limit theorem does not rescue you (small n, very heavy tails, dependent observations), and what a standard error is the standard deviation of.
Grade yourself against a reference and count only the answers that were exactly right, not the ones that were nearly right.
Rewrite the two weakest answers the following morning from memory in full sentences.

Deliverable: Ten graded answers with an honest count of exact hits.

Day 3 — Breadth pass: modelling

Take one tabular dataset end to end in 90 minutes: a leakage-safe split, a baseline that is not a model (majority class or historical mean), one regularized linear model, one gradient-boosted tree, and a single evaluation metric chosen before you look at any result.
Write why that metric fits the cost structure: precision at a fixed recall for alerting, calibration for anything feeding a price or a threshold, ranking metrics for retrieval, and note that area under the ROC curve is insensitive to class balance in a way that can flatter a rare-positive problem.
Name the leak you were most likely to introduce (an encoding fit on all rows before splitting, or a feature computed after the label's timestamp) and write the check that would have caught it.

Deliverable: A notebook whose first cell states the metric and the baseline, plus two lines on what beat what and by how much.

Day 4 — Breadth pass: product judgement

Answer three case prompts aloud at 15 minutes each, timing how long passes before you state a success metric.
For one case write the first segmentation you would run and the row counts you expect per segment, so that a tiny segment cannot quietly drive the conclusion.
Take a metric definition you did not write, from a public dashboard, a textbook, or documentation you already have open, and list every place two analysts implementing it would diverge: which rows the denominator admits, whether the unit is an account or a person, what the time window is anchored to, and what happens to data that arrives late. Then write the one question that would close the largest of those gaps.

Deliverable: Three recorded case answers plus an ambiguity list for a metric someone else defined, ending in the single question you would ask about it.

Day 5 — Depth, first area

Rank the four areas by how many bullet points in the role description each one covers, pick the top one, and spend the entire day inside it.
Work the six hardest problems you can find in that area and for each write the generalizable move you should have reached for first, rather than the answer.
Re-solve the two you failed the same evening with notes closed.

Deliverable: Six generalizable moves written as instructions to yourself, not as solutions.

Day 6 — Depth, second area, and the seam between them

Repeat the depth protocol on the second-ranked area with the same six-problem structure.
Construct one problem that requires both areas at once, for example a metric redefinition whose effect you must validate with a test whose readout you then have to query.
Solve your own combined problem end to end and note where the handoff between the two areas cost you time.

Deliverable: One combined problem, solved end to end, with the handoff failure written down.

Day 7 — Integration and re-measurement

Re-run the six prompts from day one under the same clock and compare both correctness and time.
Run a 60-minute mixed mock that moves between areas without warning, since switching cost is what breadth passes do not train.
Write the two areas you would still fail on, and the sentence you will use in the interview when you hit one of them.

Deliverable: A before-and-after score table plus a written plan for the two remaining gaps.

Getting an analysis acted on

Work that nobody used is a common and unflattering pattern in data careers, and interviewers probe for it. Have a story about an analysis that changed a decision, and be specific about how you got it in front of the person who could act. Also have one about work that went nowhere, with your reading of why.

Describe a project where you faced a significant roadblock; how did you navigate the situation to reach a conclusion?

Medium · behavioural and stakeholder questions

Approach

Close with what you would do differently, concretely.
Quantify the outcome, including what you would not claim credit for.
Name the disagreement or constraint, and how you resolved it with evidence.

Follow-up

How did you know the outcome was caused by your change?
What did you decide not to do, and why?

Disagree with a product manager about an adherence nudge

Medium · immortal time · confounding · disagreement

A product manager proposes shipping a refill-reminder feature to the whole member base, citing an internal analysis: members with 12-month proportion of days covered at or above 0.80 had 31 percent fewer admissions than members below it. The roadmap date is three weeks out and engineering is scoped. You believe the comparison builds survival and continuous coverage into the exposure definition, and is separately confounded by the kind of member who refills on time. Deliverable: what you say in the room, what you send afterwards, and the smallest study you would accept as sufficient to ship.

Approach

The probe is whether you can be specific and non-territorial at once. A disagreement that arrives as 'the analysis is flawed' costs you the room; one that names a mechanism and offers a cheaper path keeps it.
Name the two defects separately, because they have different fixes. Classifying a member as adherent over 12 months requires them to stay alive and stay covered for those 12 months, and that guaranteed event-free interval is assigned to the adherent group. Separately, members who refill on schedule differ from those who do not in ways claims never record.
Show rather than assert. Rerun the same comparison as a landmark analysis: classify adherence over days 1 to 90, start follow-up at day 91 for everyone still enrolled and event-free, and report how much of the 31 percent survives.
Separate the association claim from the intervention claim out loud. Nothing in the analysis speaks to whether a reminder changes refill behaviour, which is the actual product question and is separately testable.
Offer the shippable path: a randomised holdout on a subset with proportion of days covered as the primary endpoint, powered on the plausible effect of a reminder rather than on the 31 percent, with a stated read date.

Follow-up

The landmark rerun still shows a 20 percent difference. Do you ship?
The product manager says a holdout delays launch. How do you cost that argument?
What primary endpoint do you choose, and why not admissions?

Push back on an over-broad extract without blocking the work

Hard · minimum necessary · de-identification · governance

A researcher asks for a one-off extract covering 12,000 members: member_id, date of birth, five-digit residential postal code, full lab_result rows including value_text, and clinical note text, held on their laptop for a retrospective analysis of one lab analyte over time. They describe it as de-identified because names are excluded. A flat refusal is not an acceptable answer. Deliverable: the column list you will actually provide with a justification per retained field, the sentence explaining why free text defeats an enumerated-identifier approach, your small-cell rule for the published stratification, and what you deliver this week.

Approach

The probe is whether privacy is a design skill for you or a compliance reflex. Flat refusal and unquestioned fulfilment both fail, for opposite reasons.
Reconstruct the analysis from the research question rather than from the request. A longitudinal trend in one analyte needs a stable pseudonymous subject key, the analyte's loinc_code, value_numeric, units, the reference bounds, and a date. It does not need names, notes, or a precise birth date.
Replace each identifier with the least precise version that still answers the question: a salted per-study pseudonym instead of member_id, age in years or an age band instead of date of birth, a consistent per-subject date offset if only intervals matter, and a coarser geography than five-digit postal code.
Explain the free-text problem in one sentence. Removing an enumerated list of identifier fields does not de-identify a note, because notes carry names, relationships, employers, dates and rare clinical detail in prose, so only a reviewed extraction into structured fields or a formal statistical determination addresses it.
Set the publication rule before the analysis runs. Suppress cells below a stated threshold and suppress complements too, because a reader holding a row total and all but one cell recovers the suppressed count by subtraction.
Keep them moving. Deliver the reduced extract into a controlled environment this week rather than onto a laptop, and offer to derive any note-based variable yourself as a structured flag so they never need the prose.

Follow-up

They say the age band loses resolution their model needs. What do you do?
How do you set the suppression threshold, and who signs off on it?
The analysis later needs linking to a second data source. What changes?