How to get hired as a biostatistician in 2026-27

What a biostatistician actually does, and the credential that gates the job

A biostatistician on a clinical development team is the person who decides what question a trial can answer and how it will be answered, writes that down before anyone sees an unblinded result, and then defends the choice to clinicians, regulators and sometimes a data monitoring committee. The work is mostly specification and judgment, written into controlled documents, long before it is computation.

The artifacts are concrete and they have names. You write the statistical sections of the protocol and the sample size justification. You write or review the randomization specification that goes to the interactive response technology (IRT) vendor. You author the statistical analysis plan (SAP), which is the document that matters most, and the shells for the tables, figures and listings that follow from it. You specify the analysis datasets at the ADaM level so that the programmers can build them. You sit in the blinded data review that decides which protocol deviations exclude a patient from the per-protocol set. You produce or review the outputs for an independent data monitoring committee. After database lock you interpret the results, write the statistical sections of the clinical study report under ICH E3, contribute to the integrated summaries of safety and efficacy that go into a submission, and draft the statistical half of the answers when a health authority sends questions back.

The job it gets confused with is statistical programming. In most large organizations these are two separate career tracks with separate titles, separate managers and separate hiring processes. The statistical programmer builds SDTM and ADaM datasets, produces the tables and figures, and performs or receives independent double programming for QC. The biostatistician specifies what should be produced, reviews it, interprets it, and signs. Applying for the wrong one of these is a common and entirely avoidable mistake, because the postings read similarly to someone outside the industry and the interviews do not resemble each other at all.

At a small biotech there is no line. One statistician may write the SAP, oversee a CRO that does the programming, review every output, present to the board, and be the only person in the company who understands why the confidence interval is the shape it is. That is a different job from the same title at a large sponsor, and you should know which one you are interviewing for.

The practical entry credential for an industry biostatistics role is a master's degree in biostatistics or statistics. That is a screening rule written into job postings and applicant tracking filters, not a law, but it is applied consistently enough that trying to route around it is usually wasted effort. A typical MS in biostatistics from a school of public health runs 1.5 to 2 years full time; an MS in statistics with trials and survival coursework reads almost identically to a hiring manager. An MPH with a biostatistics concentration is a weaker signal for pharma trial work and a strong one for public health and epidemiology roles. A master's in epidemiology is not a substitute for the statistics, and a data science master's without linear models, survival analysis and longitudinal methods will get asked uncomfortable questions in the technical panel.

A PhD costs four to six years. What it buys is specific: faster progression to Principal and Director, access to methodology and complex innovative design groups (adaptive designs, Bayesian borrowing, oncology dose optimization, external control arms), credibility in regulator-facing meetings, and eligibility for statistical reviewer roles at FDA and for academic faculty appointments. It is not required for a long senior industry career, and you should ignore anyone who tells you otherwise; many Directors of Biostatistics hold an MS and a decade of studies. The honest trade is that the PhD compounds later and costs earlier.

There is no license. Nothing legally restricts the title, there is no board exam, and no registry exists. Two voluntary accreditations do exist and it is worth being clear about their weight. The American Statistical Association offers accreditation as a Professional Statistician (PStat) and a Graduate Statistician (GStat); in pharma hiring these are rarely asked for and almost never decisive. The Royal Statistical Society's Chartered Statistician (CStat) carries more recognition in the UK and parts of Europe, and is still not a gate. SAS certifications have modest value for statistical programming roles and close to none for statistician roles, where the interview tests judgment rather than syntax.

What does function as a credential, in the sense that its absence stops you, is evidence that you have produced regulated deliverables. A SAP you authored. A sample size you justified and defended. A DSMB meeting you supported, and knowing which seat you sat in. A submission you contributed to, described precisely. For a new graduate who has none of these, the substitutes that work are an industry internship, documented consulting-center hours (how many investigators, what designs, what you delivered), and a thesis described the way a study is described rather than the way a dissertation abstract is.

The coursework that actually gets interrogated in a technical panel is narrower than a transcript: linear and generalized linear models, survival analysis, longitudinal and mixed models, categorical data, design of experiments, a dedicated clinical trials course if your program has one, missing data methods, causal inference, Bayesian methods, and real computing. If your program offers a clinical trials course and a statistical consulting practicum, take both. They are the two lines on a new graduate's transcript that hiring managers read.

• Protocol statistical sections and the sample size justification, with every assumption written down
• Randomization specification: block size, stratification factors, allocation ratio, and who holds the code
• The statistical analysis plan (SAP), signed before unblinding, plus TFL shells
• ADaM-level analysis dataset specifications, including derived endpoints and analysis flags
• Blinded data review and the protocol deviation rules that define the analysis sets
• Independent data monitoring committee outputs, and the distinct roles of blinded study statistician, unblinded reporting statistician and independent committee statistician
• Clinical study report statistical sections under ICH E3, and the integrated summaries (ISS and ISE) for a submission
• Responses to health authority questions, written under a deadline measured in days
• MS in biostatistics or statistics: the practical entry credential, 1.5 to 2 years
• PhD: 4 to 6 years, buys design leadership, methodology work, regulator-facing roles and faculty or FDA reviewer eligibility
• No license, no board exam, no legally protected title anywhere in the US, UK or EU
• ASA PStat and GStat, and RSS CStat: real accreditations, not hiring gates in pharma

Where the jobs are: CRO, FSP, sponsor, biotech, device, academia and government

The same title describes at least seven quite different jobs, and choosing badly is a common cause of leaving a role inside a year.

A CRO runs studies for sponsors under contract. The large ones (IQVIA, ICON, Parexel, Fortrea, Thermo Fisher PPD, Medpace and others) hire the bulk of entry-level biostatisticians and are the realistic first job for most MS graduates. The advantages are real: you see many studies, many therapeutic areas and many sponsors in a short time, and you learn the deliverables properly because the deliverables are the product. The costs are also real: your time is billable against a utilization target, you work to each sponsor's SOPs rather than your own, and the strategic decisions are made by people who do not work at your company. If you take a CRO role, measure it by how many studies you are lead statistician on, not how many you are staffed to.

A functional service provider (FSP) arrangement badges you to a single sponsor. You use the sponsor's systems, attend the sponsor's meetings and often feel like their employee while being paid by a CRO. It trades the CRO's variety for depth and stability, and it is a common bridge into a sponsor role. Sponsors have moved a lot of work into FSP contracts over the past few years, so a growing share of what looks like sponsor work is advertised by a CRO.

A large sponsor (big pharma) runs fewer studies per statistician and goes deeper on each. You own more of the decision, you are more likely to be in the room with a regulator, there is more process and more meeting load, and compensation at the same level is typically above CRO. Sponsors mostly hire at the experienced level, which is why the standard career shape is two to four years at a CRO and then a move.

A small or mid-size biotech is a different job wearing the same name. You may be the entire statistics function. Your real skill becomes vendor oversight: specifying, reviewing and challenging work a CRO performs, while retaining accountability for it. You present to a board. A single readout can determine whether the company continues to exist. The upside is ownership and equity; the risk is that there is nobody senior to check your work, so take one only when you can show someone experienced is available to you, even as an external advisor.

Medical device and in-vitro diagnostics is a genuinely distinct statistical culture. Sample sizes are smaller, non-inferiority and agreement studies are everywhere, sensitivity and specificity and ROC analysis replace time-to-event as the daily currency, Bayesian designs have been accepted by FDA's device center for longer than on the drug side, and the governing standard for clinical investigations is ISO 14155 alongside the IDE regulation rather than the drug trial apparatus. Statisticians who move in from pharma usually enjoy it and find the first six months disorienting.

Academic medical centers hire biostatisticians into collaborative cores on grant funding. You get breadth (one week a surgical outcomes study, the next a trial), co-authorship, and often an academic title. Pay is materially lower than industry at the same experience level and the funding is less stable. Government and public health work (CDC, NIH and the NCI, state health departments) sits in the middle: slower, more mission, pension-shaped benefits, and most federal posts require US citizenship, so read the eligibility line before you spend a day on the application. FDA statistical reviewer roles in CDER's Office of Biostatistics are a serious and underrated career path, usually requiring a PhD, and former reviewers are heavily recruited by industry afterwards.

Two practical notes on the 2026-27 market. First, it is tighter than the hiring surge of 2020 to 2022: biotech funding contracted, several sponsors restructured, and new graduates are competing against experienced statisticians who were displaced. That changes tactics rather than prospects, because the work itself has not gone away, but it means applying wider than big pharma and treating CRO, FSP, device, RWE, government and academic cores as the same search. Second, therapeutic area is leverage. Oncology is the largest single area of industry trial activity, which makes fluent time-to-event analysis the most portable skill on your resume, and vaccines, rare disease and cell and gene therapy each hire for their own quirks (immunogenicity endpoints, tiny samples and external controls, long-term follow-up).

• CRO: most entry-level roles, broad exposure, utilization targets, sponsor-driven SOPs
• FSP: badged to one sponsor, their systems and SOPs, depth over variety, a common bridge
• Large sponsor: fewer and deeper studies, regulatory interaction, higher pay, hires mostly at experienced level
• Small biotech: you are the function, the skill is CRO oversight, equity and single-readout risk
• Device and diagnostics: ISO 14155, agreement and accuracy studies, Bayesian designs more established
• Academic core: breadth and authorship, grant funding, lower pay, usually PhD for faculty and MS for staff
• Government and FDA: CDC, NIH, and CDER statistical reviewer roles that are a strong launchpad back into industry, with citizenship requirements on most federal posts
• Market in 2026-27: tighter than the 2020-2022 surge, more FSP-badged roles, oncology still the biggest pool of work

SAS versus R in 2026-27, answered properly

This is the question searchers ask most often about this role, and most of the answers online are a decade out of date in one direction or wishful thinking in the other.

Start with what is true and frequently misstated: FDA does not require SAS. There is no regulation naming a statistical package. What FDA's study data technical conformance guide asks for is study datasets submitted in SAS Transport format (XPT version 5), which is a file format specification that other tools can write, documented by define.xml. The widely repeated belief that a submission must be produced in SAS is wrong, and stating it confidently in an interview will mark you as someone who repeats industry folklore. The same guide also asks for the programs that create the ADaM datasets and the primary and secondary efficacy outputs, which is why reproducible code is a regulatory expectation rather than a personal habit. Check the current version of the guide rather than quoting a remembered one.

Now what is also true: SAS remains the production language for regulated deliverables at most large sponsors and most CROs, and will be for the period you are planning around. The reasons are structural rather than regulatory. Validated computing environments are expensive to qualify and slow to change. Twenty years of macro libraries encode a company's conventions for every table shell it has ever shipped. Double programming for QC is built around SAS staffing. And a submission has to be reproducible years later by people who did not write it. None of that evaporates because a better language exists.

What has genuinely changed is that R crossed from 'fine for exploratory work' to a first-class submission language at a growing number of organizations. The R Consortium's R Submission Working Group has run a sequence of pilot submissions to FDA demonstrating R-based analysis packages, and more recently container and web-application based submissions, and FDA engaged with them. The pharmaverse ecosystem made this practical rather than theoretical: admiral for building ADaM datasets, rtables and tern for regulatory-shaped tables, Tplyr, gtsummary for summary tables, and the R Validation Hub's risk-based framework for assessing whether a package is fit for regulated use. Biotechs founded in the last several years are frequently R-first by default, because they never accumulated the SAS estate in the first place.

So the correct answer to 'SAS or R' in an interview is neither tribal nor evasive. It is: production deliverables in whichever the sponsor's validated environment requires, which today is usually SAS at a large organization and increasingly R at a small one; R for simulation, modeling, graphics and exploratory work almost everywhere; and here is specifically what I have built in each. Then be concrete. A candidate who says 'proficient in SAS and R' and cannot name the procedure they would use for a mixed model for repeated measures has told the panel something they did not intend to.

Two more things belong in this answer. First, reproducibility tooling is a differentiator rather than a nice extra, because it is increasingly inspected: version control with Git, environment pinning with renv or a container, a pipeline tool such as targets, and a clear statement of what independent double programming means in your workflow and who signs off. Second, the sample size and design software is its own skill set and is frequently tested: East (Cytel) and nQuery for power and group sequential designs, FACTS for adaptive designs, PROC POWER, PROC SEQDESIGN and PROC SEQTEST in SAS, and Stan or JAGS when the design is Bayesian.

• FDA requires SAS Transport (XPT v5) as a dataset file format, not SAS as software. Know the difference and say it correctly
• Submissions also carry the programs that build ADaM and the key efficacy outputs, plus define.xml. Reproducibility is a regulatory expectation
• SAS procedures to actually know: MIXED and GLIMMIX for MMRM, LIFETEST and PHREG for survival, LOGISTIC, FREQ with CMH and exact options, MI and MIANALYZE, POWER, SEQDESIGN and SEQTEST, plus the macro language and ODS
• R packages to actually know: mmrm, lme4 and nlme, survival, emmeans, brms or rstanarm, admiral, rtables and tern, gtsummary, plus renv and targets for reproducibility
• Design software: East (Cytel), nQuery, FACTS, PASS. Naming the one you used and the assumptions you entered is worth more than listing all four
• Python appears for simulation, pharmacometrics adjacency and machine-learning-flavored work, rarely for regulated deliverables
• Git literacy is now assumed at biotechs and increasingly asked about at sponsors

The resume a biostatistics hiring manager actually reads

Two pages for industry. A long academic CV is correct for a faculty application and wrong for a sponsor or CRO, where the hiring manager is scanning for therapeutic areas, phases, deliverables and whether you were lead or support.

The unit of a biostatistics resume is the study, not the duty. A bullet that says 'performed statistical analyses for clinical trials' conveys nothing, because every applicant could write it. Write each significant study as: phase, indication, design, size, allocation and stratification, primary endpoint, the estimand and primary analysis method, your role, the deliverables you personally produced, and what happened next.

Here is the shape, with the detail level that works. The study below is invented; the structure is the point. 'Lead statistician, phase 3 randomized double-blind placebo-controlled trial in moderate to severe atopic dermatitis (N=620, 2:1, stratified by prior systemic therapy and baseline severity). Authored the SAP and TFL shells. Primary estimand used a treatment-policy strategy for rescue medication; primary analysis logistic regression with multiple imputation; two interim analyses with a Lan-DeMets O'Brien-Fleming alpha spending function. Produced unblinded DSMB outputs through an independent reporting group, supported the BLA statistical sections, and drafted the statistical response to two FDA information requests.' A hiring manager can price that paragraph in ten seconds.

Be exact about submission and committee claims, because the follow-up questions are precise and overclaiming is caught in the first panel. 'Supported the ISS' is different from 'led the ISS'. 'Study statistician, blinded throughout' is a different job from 'unblinded reporting statistician for the IDMC' and from 'independent statistician to the committee'. Conflating those three reads as inexperience even when the rest of your record is strong. Say which seat you were in.

What gets ignored or actively hurts: skills rated as bars or percentages, 'excellent communication skills' asserted rather than demonstrated, GPA after your first job, a list of every SAS procedure you have ever run, forty publications on an industry resume (choose three to five that show methods relevant to the role), coursework listed by an experienced hire, a photograph, and an objective statement. Also avoid the phrase 'strong analytical skills' entirely; it is one of the most common lines on a statistician's resume and the least informative.

For a new graduate with no studies, the substitutes work if you write them the same way. Translate the thesis into a study-shaped bullet: the question, the design or data source, the sample, the method, what you concluded, and where it was presented or published. Count your consulting practicum work: how many investigators, what designs, what you delivered and in what software. List an internship with the deliverables rather than the company name alone. A public, reproducible analysis in a Git repository with a readme that explains the design decisions is worth more than another course certificate, because it is the only thing on the page the panel can inspect.

On applicant tracking systems: the first screen for a CRO role is often keyword-based, so the vocabulary of the posting needs to appear in your resume where it is true. If the posting says ADaM, estimand, MMRM, oncology and SDTM, and you have done those things, use those words rather than paraphrases. Spell them the way the posting spells them, including randomization or randomisation, because a literal keyword match is literal. Do not pad with terms you cannot defend, because the first technical conversation will test them.

• Two pages, study-level bullets, phases and therapeutic areas visible in the top third
• For each study: phase, indication, design, N, randomization and stratification, primary endpoint, estimand and analysis method, your role, your deliverables
• Name the deliverables you authored: SAP, TFL shells, randomization spec, ADaM specifications, CSR sections, ISS or ISE contributions, health authority responses
• State your DSMB or IDMC seat precisely: blinded study statistician, unblinded reporting statistician, or independent committee statistician
• Cut: skills bars, GPA after the first job, every PROC you know, forty publications, coursework for experienced hires, 'strong analytical skills'
• New graduates: thesis as a study, consulting practicum counted in clients and designs, internship by deliverables, one inspectable reproducible repo
• Mirror the posting's vocabulary and spelling where it is true, because the first CRO screen is often keyword matching

How the hiring process actually runs, stage by stage

Hiring for this role is a structured multi-stage process almost everywhere except the smallest biotechs, and it is slower than most candidates expect.

The first screen is an internal recruiter, usually 20 to 30 minutes, filtering on degree level, years of experience, therapeutic areas, software, work authorization and location. For remote roles the real constraint is often time zone overlap with a study team rather than geography. Recruiters at CROs frequently screen for a specific requisition tied to a specific sponsor contract, so asking which study and which sponsor is a fair question that also tells you how real the role is.

The second stage is the hiring manager, typically an Associate Director or Director of Biostatistics, for 45 to 60 minutes. The design conversation normally starts here, in a light form: tell me about a study you led, what the primary endpoint was, why that analysis. People who treat this as the friendly introduction before the technical rounds get caught out.

The third stage is a panel, usually two to four conversations of 45 to 60 minutes, sometimes compressed into a half day. The panel is deliberately mixed. Another statistician tests methods and design judgment. A statistical programming lead tests whether you can specify what they will have to build, which means ADaM literacy, derivation clarity and whether you understand what makes a spec ambiguous. A clinical or medical person tests whether you can explain a result to someone who will act on it. Sometimes data management or regulatory joins to test cross-functional working.

For PhD and senior hires a presentation is common: 20 to 30 minutes on a study you ran or your thesis, followed by questions. The questions are about your choices, not your results. Why that endpoint, why that covariance structure, what you would do differently, what you would have done if the assumption had failed. Prepare the version of the talk where every slide can survive 'why'.

Some CROs use a written statistical test or a short SAS or R exercise, and a few use a take-home. Ask before accepting what it covers and how long it should take; an unbounded take-home for a mid-level role is a reasonable thing to push back on. The exercises are usually not exotic: derive an analysis variable, run a model, produce a table in the right shape, interpret an odd result, or find the bug in some code.

Expect two to five weeks end to end at a CRO, six to twelve at a large sponsor where headcount approvals and panel scheduling dominate, and occasionally under a week at a biotech that needs cover before a readout. Offers are followed by degree verification and reference checks, and sometimes transcripts, so do not round a degree up on a resume.

If you need visa sponsorship, deal with it in the first conversation rather than the last. Biostatistics cohorts in the US are heavily international, statistics degrees normally qualify as STEM for the extended post-study work period (confirm the current rules with your university's international office rather than a forum), large CROs and large sponsors sponsor routinely, small biotechs often will not, and most federal roles including FDA reviewer posts require US citizenship. Recruiters are not offended by the question; they are staffing a requisition with a budget line for it or they are not.

One route that is underused: conferences in this field are genuine hiring venues, not just talks. The ASA Joint Statistical Meetings runs a career placement service where sponsors and CROs interview on site. The Regulatory-Industry Statistics Workshop puts you in a room with FDA statisticians and industry leads. ENAR's Spring Meeting draws US biostatistics departments and the recruiters who follow them, and PSI's annual conference plays a similar role in Europe. For a new graduate, one of these is worth more than two hundred online applications.

• Recruiter screen, 20 to 30 minutes: degree, years, therapeutic area, software, authorization, time zone
• Hiring manager, 45 to 60 minutes: the design conversation starts here, not later
• Panel of two to four: a statistician for methods, a programming lead for specification clarity, a clinician for communication, sometimes data management or regulatory
• Presentation for senior and PhD hires: 20 to 30 minutes, questioned on choices rather than findings
• Optional technical exercise at some CROs: a short stat test, a SAS or R task, occasionally a take-home. Ask the scope before accepting
• Timeline: 2 to 5 weeks at a CRO, 6 to 12 at a large sponsor, days at a biotech under readout pressure
• Visa sponsorship: raise it in the recruiter screen, expect large CROs and sponsors to sponsor and most federal roles to require citizenship
• Conferences that really hire: JSM career placement, the Regulatory-Industry Statistics Workshop, ENAR Spring Meeting, PSI in Europe

The interview: trial design, estimands and the whiteboard sample size

Almost every technical interview for this role is a trial design conversation wearing different clothes. Whether they open with 'design a phase 3 for this indication', 'walk me through your last study', or 'what would you do if the proportional hazards assumption failed', they are testing the same thing: can you make and defend a sequence of design decisions under uncertainty, and do you know when to say the trial cannot answer the question being asked.

Answer design questions with a structure, out loud, and start with the decision rather than the method. What decision does this trial support, and for whom: a go or no go internally, a regulatory claim, a label expansion, a reimbursement dossier? Then: population and key eligibility, endpoint, the estimand with its intercurrent events, choice of control and whether placebo is ethical here, randomization ratio and stratification factors, blinding and who stays unblinded, sample size with every assumption stated aloud, multiplicity across endpoints and doses, interim analyses and stopping rules, missing data handling and the sensitivity analyses that probe it, prespecified subgroups and what you will and will not claim from them, and the monitoring committee. Finish with the conditions under which you would advise against running it. Candidates who begin by naming a test rather than a decision lose the room in the first minute.

Estimands are the single most reliable interview topic in this field now, and the drill is always a scenario. A patient discontinues treatment because of toxicity. A patient starts rescue medication. A patient dies before the endpoint can be measured. A patient is lost to follow-up. The expected answer: name the intercurrent event, name the strategy you would use (treatment policy, hypothetical, composite variable, while on treatment, or principal stratum), explain what the resulting quantity means clinically, and say plainly that the choice belongs to the clinical and regulatory team with the statistician framing the options and their consequences. Reciting the five attributes of an estimand with no scenario attached is the most common way strong candidates sound junior.

Expect to produce a sample size without software. The rule of thumb worth having memorized for a two-arm comparison of means is roughly 16 divided by the square of the standardized difference per arm for 80% power at two-sided alpha 0.05, and roughly 21 over the same quantity for 90% power. Say the assumptions out loud, inflate for dropout, state the effect you consider clinically meaningful rather than the one that makes the number small, and then say you would confirm in nQuery, East or PROC POWER and run the operating characteristics. Being unable to produce an approximate number without a tool is a real and frequently fatal weakness in this interview.

On missing data, know the assumptions and not just the procedures. Be able to distinguish missing at random from missing not at random and say which one your primary analysis assumes. A mixed model for repeated measures with an unstructured covariance matrix and a Kenward-Roger or similar degrees-of-freedom adjustment is the usual primary for a continuous longitudinal endpoint; multiple imputation is the usual alternative; control-based imputation, jump-to-reference and tipping point analyses are the sensitivity analyses that test departures from MAR. Last observation carried forward as a primary analysis is a wrong answer and has been for many years.

On time-to-event, expect the proportional hazards assumption to be attacked. Know how you would check it, and what you would do if it fails: restricted mean survival time, a weighted log-rank test, milestone survival at a prespecified time, or a different estimand entirely. Know the difference between a cause-specific hazard model and a Fine-Gray subdistribution model, and what each one actually estimates, because this is the question that separates people who learned survival analysis from people who have used it. If you are interviewing in oncology, which is the largest pool of industry work, also be ready on censoring rules, blinded independent central review versus investigator assessment, and what informative censoring does to a progression-free survival estimate.

On multiplicity, be able to build a testing strategy rather than name a correction. Hierarchical and fixed-sequence testing, Holm and Hochberg, graphical approaches for structured hypotheses, and gatekeeping across families of doses and endpoints. Say what the Type I error is being controlled for: the claims that will appear in the label. On non-inferiority, be able to justify a margin from historical data and explain the constancy assumption and assay sensitivity, and why a non-inferiority trial is harder to defend than a superiority trial. On Bayesian methods, be able to say when you would use them, how you would justify a prior, how borrowing works (MAP priors, power priors, commensurate priors), and that frequentist operating characteristics still have to be simulated and shown to a regulator. On adaptive designs, distinguish blinded from unblinded sample size re-estimation, know why unblinded adaptation threatens Type I error and how combination tests or conditional error functions preserve it, and say that any non-standard design needs prespecification and a simulation report.

Know the current shape of the guidance landscape, and quote it carefully. GCP has been revised, and ICH E6(R3) is the revision to name if someone asks you about GCP; regional adoption and transition timing vary, so check your region rather than asserting a date. An ICH guideline on adaptive designs, E20, has been moving through the ICH process, so know it exists and check what step it has reached before citing it as settled. In oncology, FDA's Project Optimus changed the dose question from 'what is the maximum tolerated dose' to 'why is this the optimal dose', which is why randomized dose comparison in early development is now a live interview topic and why being able to compare 3+3 with model-assisted designs such as BOIN or mTPI and model-based CRM is worth more than it used to be. Saying 'I would check the current version of that guidance' costs you nothing and is the correct professional instinct; misquoting a date in a regulatory interview is expensive.

There is also a communication test, and it is not a formality. Someone, usually the clinician on the panel, will ask you to explain a p-value, a confidence interval or a hazard ratio to a non-statistician, or to deliver a negative result to a team that wanted a positive one. Practice this aloud before the interview. Technically excellent candidates fail here more often than they fail on methods, usually by being either too long or faintly condescending. The strongest answer is short, uses the trial's own units, and ends with what the team should do next.

Finally, ask questions that reveal the job. How many studies will I be lead statistician on in the first year? Who authors the SAP on your studies? What is the programming support model and the ratio? How much of the role is sponsor-facing or regulator-facing? What was the last submission this group supported? For a CRO: what is the utilization target? The answers to those six will tell you more than the job description did.

• Open with the decision the trial supports, then population, endpoint, estimand, control, randomization, blinding, sample size, multiplicity, interims, missing data, subgroups, monitoring committee
• Intercurrent event strategies to name on demand: treatment policy, hypothetical, composite variable, while on treatment, principal stratum
• Whiteboard sample size: about 16 over the squared standardized difference per arm for 80% power at two-sided 0.05, about 21 for 90%, then inflate for dropout and confirm in software
• Missing data: state MAR versus MNAR, MMRM or multiple imputation as primary, control-based imputation and tipping point as sensitivity. LOCF as primary is a wrong answer
• Survival: how you check proportional hazards and what you do when it fails (RMST, weighted log-rank, milestone), cause-specific versus Fine-Gray, censoring rules and central review in oncology
• Multiplicity: hierarchical and fixed-sequence testing, Holm and Hochberg, graphical approaches, gatekeeping, and what the Type I error protects
• Non-inferiority: margin justification, constancy, assay sensitivity, and analyzing both ITT and per-protocol
• Bayesian and adaptive: prior justification, borrowing methods, blinded versus unblinded sample size re-estimation, and simulated operating characteristics every time
• Oncology dose finding after Project Optimus: 3+3 versus BOIN or mTPI versus CRM, and randomized dose comparison before the phase 3 dose is fixed
• Guidance hygiene: name ICH E9(R1), E3, E6(R3) and E17 confidently, and say you would check the current status rather than quoting a date you half remember
• Communication: explain a hazard ratio in the trial's own units in under a minute, and be ready to deliver a negative result well

Pay, levels and the parts of the offer that decide whether you stay

Be skeptical of any single salary number for this role, including ones you find on aggregator sites, because the spread across CRO, sponsor, biotech, academia and geography is wider than the differences the aggregators model. Use sources you can check. BLS Occupational Employment and Wage Statistics code 15-2041 covers Statisticians and gives you national and metropolitan wage distributions; the detailed O*NET occupation 15-2041.01 is Biostatisticians specifically. Several US states require employers to publish a pay range in the posting, Colorado, California, Washington, New York and Illinois among them, which means you can read actual employer ranges for actual requisitions rather than a crowd-sourced average. The American Statistical Association publishes salary survey work for the profession; check how recent the edition is before you lean on it. For academic posts, public university salary disclosures and AAMC faculty salary data are the right sources.

What is reliable is the shape rather than the number. Compensation rises with degree level, with years, and above all with proximity to the decision: the statistician who owns a program and talks to regulators is paid differently from the one who executes a study well. Sponsors generally pay above CROs at the same level. Small biotechs trade cash for equity and should be evaluated on the cash, with the equity treated as a lottery ticket you can afford to lose. Academic medical centers pay materially less and compensate in variety, authorship and autonomy. Within the US, the Boston, San Francisco Bay Area, New Jersey, Philadelphia and Research Triangle clusters price higher, though remote hiring has flattened this more for biostatistics than for most pharma roles.

The ladder is reasonably standard: Biostatistician I and II, Senior Biostatistician, Principal Biostatistician or Statistical Scientist, Associate Director, Director, Senior or Executive Director, and Vice President of Biostatistics. The step that changes the job rather than the salary is Associate Director at a sponsor, which is usually where you start owning a program and its regulatory strategy instead of a study and its outputs. A parallel non-managerial track exists at many companies through Principal and Fellow, and it is worth asking whether it is real at this employer or decorative.

When you have an offer, the terms that determine whether you last two years or ten are mostly not financial. How many studies will you be lead on, and how many will you merely support? Will you author SAPs or review them? Will you present to a data monitoring committee or to a health authority, and when? What is the programming support ratio, and does the programming sit in the same organization as you? At a CRO, what is the utilization target, and how much non-billable time exists for method work and training? If the role is remote, is the team distributed or are you the only person not in the building, because the second one is a different experience.

Three red flags worth naming. A sponsor-facing role where you will never author a SAP is a reviewing job with a doing job's title. A CRO utilization target so high that there is no non-billable time means you will not develop. And a biotech where you will be the only statistician with no external statistical advisor and no access to a consultant is a role where your first serious mistake will be expensive and unobserved. None of these is disqualifying on its own. All of them are worth asking about before you sign.

• Pay sources worth citing: BLS OES 15-2041, O*NET 15-2041.01, state pay-transparency ranges in live postings, ASA salary survey work, AAMC and public university data for academia
• Ladder: Biostatistician I and II, Senior, Principal or Statistical Scientist, Associate Director, Director, Senior or Executive Director, VP
• Associate Director at a sponsor is the step where you start owning a program rather than a study
• Offer questions that matter: studies led versus supported, SAP authorship, DSMB and regulator exposure, programming support ratio, CRO utilization target
• Red flags: a senior title with no SAP authorship, a utilization target with no non-billable time, a lone statistician at a biotech with no external advisor

Breaking in with no industry experience: what to do over the next 90 days

Most advice for new graduates stops at 'get experience'. Here is the version you can actually start on Monday, in the order that pays off fastest.

First, pick the track and stop hedging. Statistician or statistical programmer. Read ten live postings from CROs for each and list the deliverables; the posting that says 'author the SAP, justify the sample size, interpret results' is the statistician job and the one that says 'build SDTM and ADaM, produce TFLs' is the programming job. Applying to both with the same resume is how people get neither. Programming is an honest deliberate route in if the statistician screen keeps rejecting you, and plenty of senior statisticians started there.

Second, learn the deliverables from real documents, which are free. Large trials published in major medical journals routinely post the full protocol and the statistical analysis plan as supplementary material, and some sponsors publish them on their own trial transparency pages. Take one, read the SAP end to end, and then write your own TFL shells for its primary endpoint and your own ADaM specification for the analysis flag it depends on. Nothing else you can do in a week teaches you as much about what the job is, and it gives you something true to say when a panel asks whether you have seen a SAP.

Third, build one inspectable artifact instead of three certificates. A small simulation study is the best choice because it rehearses the interview: simulate a two-arm trial, implement a group sequential boundary, and report Type I error, power and expected sample size across a grid of scenarios including the ones where the design behaves badly. Put it in a Git repository with renv or a container, a readme that explains the design decisions rather than the file layout, and a short results table. That is the only thing on your application a hiring manager can open and check.

Fourth, read ICH E9 and the E9(R1) addendum properly. Both are short enough to read in a couple of evenings, they are the source of the questions you will be asked, and almost no new graduate has actually read them rather than a summary of them.

Fifth, work the real channels rather than the aggregators. Apply directly on CRO career sites, where Biostatistician I requisitions appear and where the keyword screen is the first gate. Ask your department which sponsors and CROs recruit from your program, because recurring relationships place more graduates than cold applications do. Register for ENAR's Spring Meeting or JSM and use the career placement service rather than attending only the talks. Ask faculty for consulting-center hours now, even unpaid, and record every engagement as clients, designs, deliverables and software.

Sixth, rehearse out loud. The whiteboard sample size, an estimand scenario, a hazard ratio explained to a clinician in under a minute, and a two-minute walkthrough of your thesis as a study rather than as a dissertation. These are the four things the panel will actually ask for, and the gap between knowing them and being able to say them under mild pressure is larger than people expect.

• Week 1: pick statistician or programmer by reading twenty live postings and listing their deliverables
• Weeks 1 to 3: read a published protocol and SAP from a journal supplement, then write TFL shells and an ADaM spec for its primary endpoint
• Weeks 2 to 6: build one simulation repository with renv or a container, a readme explaining decisions, and operating characteristics across scenarios
• Weeks 2 to 4: read ICH E9 and E9(R1) in full, not a summary
• Ongoing: apply directly on CRO career sites, ask your department who recruits there, register for ENAR or JSM and use the career service
• Ongoing: consulting-center hours recorded as clients, designs, deliverables and software
• Before any interview: rehearse the whiteboard sample size, an estimand scenario, a hazard ratio in plain words, and your thesis as a study

What a screen is looking for

These are the terms that a resume screen, human or automated, is matching against for this role. Use the ones that are true of you, in the words the posting uses.

• Biostatistician
• Biostatistics
• Clinical trial statistician
• Statistical Scientist
• Senior Biostatistician
• Principal Biostatistician
• Associate Director Biostatistics
• Biostatistician I
• Statistical programming
• Statistical Analysis Plan (SAP)
• Protocol statistical section
• Sample size calculation
• Power calculation
• Randomization specification
• Randomisation specification
• Stratified randomization
• Stratified randomisation
• Estimand
• ICH E9(R1)
• ICH E9
• Intercurrent events
• Treatment policy strategy
• Hypothetical strategy
• Composite variable strategy
• While on treatment strategy
• Principal stratum strategy
• ICH E3
• ICH E6(R3)
• ICH E10
• ICH E17
• ICH E20 adaptive designs
• Good Clinical Practice
• 21 CFR Part 11
• SDTM
• ADaM
• ADSL
• ADAE
• ADTTE
• ADLB
• define.xml
• Analysis Results Metadata
• CDISC
• Controlled terminology
• Tables Figures and Listings (TFL)
• TLF shells
• Double programming
• Independent QC
• Clinical Study Report
• Integrated Summary of Safety (ISS)
• Integrated Summary of Efficacy (ISE)
• NDA
• BLA
• MAA
• IND
• Pre-NDA meeting
• Health authority response
• FDA statistical review
• DSMB
• IDMC
• Unblinded reporting statistician
• Blinded data review
• Protocol deviation
• Analysis sets
• Intent-to-treat
• Per-protocol
• SAS
• SAS/STAT
• PROC MIXED
• PROC GLIMMIX
• PROC PHREG
• PROC LIFETEST
• PROC LOGISTIC
• PROC FREQ
• PROC MI
• PROC MIANALYZE
• PROC POWER
• PROC SEQDESIGN
• PROC SEQTEST
• SAS macro language
• SAS Transport XPT
• R
• pharmaverse
• admiral
• rtables
• tern
• gtsummary
• Tplyr
• mmrm package
• survival package
• emmeans
• lme4
• brms
• Stan
• renv
• targets
• Git
• Python
• East (Cytel)
• nQuery
• FACTS
• PASS
• MMRM
• Mixed model for repeated measures
• Multiple imputation
• Control-based imputation
• Tipping point analysis
• Missing at random
• Missing not at random
• Kaplan-Meier
• Cox proportional hazards
• Proportional hazards assumption
• Restricted mean survival time (RMST)
• Weighted log-rank
• Competing risks
• Fine-Gray
• Logistic regression
• ANCOVA
• Generalized linear models
• Longitudinal data analysis
• Bayesian methods
• MAP prior
• Power prior
• Commensurate prior
• Group sequential design
• Alpha spending
• O'Brien-Fleming
• Lan-DeMets
• Interim analysis
• Futility analysis
• Sample size re-estimation
• Adaptive design
• Master protocol
• Basket trial
• Platform trial
• Non-inferiority margin
• Multiplicity adjustment
• Hochberg
• Holm
• Graphical multiple testing
• Gatekeeping
• Subgroup analysis
• Covariate adjustment
• Simulation study
• Operating characteristics
• Oncology
• Project Optimus
• Dose optimization
• Dose escalation
• BOIN design
• Continual reassessment method
• RECIST
• Blinded independent central review
• Progression-free survival
• Overall survival
• Vaccines
• Rare disease
• Cell and gene therapy
• Medical device statistics
• ISO 14155
• Sensitivity and specificity
• ROC analysis
• Agreement study
• Digital health technology endpoints
• Real-world evidence
• External control arm
• Target trial emulation
• Propensity score
• Health technology assessment
• Indirect treatment comparison
• CRO
• Functional service provider (FSP)
• Sponsor oversight
• Vendor oversight
• MS Biostatistics
• PhD Biostatistics
• BLS OES 15-2041

Mistakes that cost people this job

Questions people ask