RESEARCH DISCIPLINE
Outcomes Research Statistics
Effectiveness in Routine Practice
For comparative effectiveness and real-world evidence studies: a defined cohort and time zero, confounding by indication addressed, and effect estimates with confidence intervals and bias analyses that a reviewer or payer can examine.
No Commitment Required
Strictly Confidential
In-House PhD-Level Statisticians
Fixed Project Fee
Same-Day Response
What Does Statistical Analysis in Outcomes Research Involve?
Outcomes research statistics estimates how treatments, programs and care pathways perform in routine practice, usually from claims, registry or electronic health record data in which treatment was not randomized. It covers the definition of the cohort and of time zero, control of confounding by indication, the choice of an outcome model and an assessment of how far unmeasured confounding could change the conclusion. OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, supports investigators and graduate researchers in outcomes research: an in-house PhD-level statistician builds the analysis on your data and states what the design can and cannot support. Which comparison to make, and what the result means for patients and payers, remain your decisions.
WHEN TO ENGAGE
Where Effectiveness Studies Need Statistical Judgment
Questions about confounding and timing begin with the first protocol draft and continue to the reviewer's reply.
Protocol and Time Zero The entry rules, the comparator and the start of follow-up are defined before data are pulled, so that treatment groups are comparable at baseline and no person-time is misclassified.
Database in Hand A claims or registry extract exists, and the inclusion criteria and the coding of exposure, outcome and covariates must be settled before any treatment effect is estimated.
Treatment Effect to Estimate Two or more treatments are to be compared on an outcome, with confounding addressed by matching, weighting or modeling, and the estimand named so that the result answers a clear question.
Reviewer or Payer Challenge A reviewer, payer or committee asks how unmeasured confounding or the choice of comparator affects the result, and the analysis is extended with evidence and not defended by assertion.
WHAT THE SUPPORT COVERS
Statistical Components of Outcomes Studies We Support
The work divides into six parts, from the trial being emulated to the reporting checklist, and a study may need all of them or only the later ones.
Design and Estimand
A comparative effectiveness research analysis begins from the trial it would emulate, with eligibility, treatment strategies, time zero and follow-up written down before any outcome is examined.
Cohort Construction
New-user cohorts, washout periods and exposure definitions are built from the source data, and each exclusion is recorded with its reason and its effect on the sample.
Confounding Control
Confounding by indication is addressed through Propensity Score Matching (PSM), weighting or outcome models, with balance and overlap examined.
Policy and Program Effects
Where a coverage change or program began on a known date, the effect is studied with Interrupted Time Series, with the earlier trend examined.
Resource Use and Cost
Cost and utilization outcomes are modeled with methods that suit skewed data, following the conventions of Health Economics research.
Reporting Routine-Data Studies
Findings are reported with estimates and intervals under the RECORD extension of STROBE, as expected in Health Sciences research.
COMMON ISSUES
Biases That Routine Data Introduce
Real-world evidence analysis inherits biases that randomization would have removed, and each one below is either designed out or bounded in a sensitivity analysis before a treatment effect is stated.
Immortal Time Built In Follow-up starts before treatment begins, so patients must survive to be treated and the treated group seems to fare better, a bias avoided when eligibility, the start of treatment and the start of follow-up coincide, as in a target trial emulation.
Prevalent Users Included Patients already on a drug are compared with new starters, which selects those who tolerated it early and distorts the comparison unless a new-user design is used.
Balance Assumed Propensity scores are estimated and the effect is reported without checking whether the weighted groups are alike or overlap, though poor overlap leaves some patients with no comparable counterpart.
Unmeasured Confounding Ignored Only the recorded covariates are adjusted for, and no bias analysis shows how strong an unmeasured factor would have to be to change the conclusion.
Outcomes Misclassified Diagnosis codes stand in for clinical outcomes without a check of how well the code identifies the condition, so misclassification can move the estimate in either direction.
Causal Wording Beyond the Design A difference between two drugs in claims data is described as the effect of the drug, when the design supports an adjusted association unless its assumptions are argued and tested.
DELIVERABLES
What You Receive for Your Effectiveness Study
Each file answers a question a reviewer or payer may ask about the comparison, from how the cohort was built to how strong an unmeasured confounder would need to be.
Study Protocol Document A written protocol with the estimand, cohort definitions, covariates, model specification and prespecified sensitivity analyses, which can be registered or placed in an appendix.
Analytic Cohort and Flow Diagram The analytic dataset with variable definitions and a flow diagram showing how many patients each eligibility rule removed, so that the final sample can be traced.
Balance and Overlap Report Standardized differences before and after adjustment, with overlap plots of the propensity score, so that a reader can judge whether the groups are comparable.
Effect Estimate Tables Hazard ratios, risk differences or mean differences as the estimand requires, each with its 95% confidence interval and the number of patients behind it.
Sensitivity and Bias Analyses Alternative exposure and outcome definitions and a quantitative bias analysis, showing how far the conclusion depends on each analytic choice.
Data Limitations Statement A note on what the claims or registry data cannot capture, matched to the output, for you to check and incorporate, with the analysis code attached.
ILLUSTRATIVE EXAMPLE
Two Diabetes Drugs in a Claims Database
An illustration of how a comparative effectiveness question is analyzed when treatment was not randomized.
ILLUSTRATIVE EXAMPLE · HEALTH INSURANCE CLAIMS RESEARCH · UNITED STATES
A health services researcher wanted to compare cardiovascular hospitalization in adults with type 2 diabetes who started one of two second-line drugs, using several years of insurance claims.
Treatment was not randomized, so the protocol was written first: a new-user cohort with a washout period, time zero at the first prescription fill, a prespecified covariate set and a named estimand. The difficulty lay in the groups themselves: they differed in age and prior cardiovascular disease, and some patients had no comparable counterpart in the other group. Propensity score weighting was used with balance checked before any outcome was examined, patients outside the region of overlap were trimmed by a stated rule, and a quantitative bias analysis asked how strong an unmeasured confounder would need to be. The finding was described as an adjusted association between drug and outcome in routine care, with the limits of claims data stated.
Outcome: The researcher received the weighted hazard ratio with its 95% confidence interval, the balance and overlap report, the bias analysis, the cohort flow diagram and the syntax for the paper.
This example does not describe a real client or project. We never share details of actual client work, in line with the confidentiality we promise every researcher. This illustrative example is included to show the type, scope, and standard of work involved in this kind of analysis.
SOFTWARE
Analysis Environments for Routine Data
Claims extracts often arrive as SAS files while weighting and bias analyses may be scripted in R or Stata, and the choice follows where your data sit and what your coauthors can rerun.
HOW WE ASSESS YOUR PROJECT
We Review Your Actual Research
Your assessment is based on a review of your actual research materials and analytical requirements, not a generic project category.
You Submit Your Project Details
Share whatever is available at the current stage, such as the research aim, protocol or proposal, dataset, variable list, manuscript or statistical requirements.
We Review Your Research and Define the Scope
Your statistically trained Research Coordinator reviews the materials to understand the study design, research questions, data structure, analytical requirements and expected deliverables. If clarification is needed, additional documents or information can be requested before the assessment is finalized.
Your Analysis Is Completed and Delivered
Once the scope and fixed project fee are agreed, the in-house statistician completes the work and the Research Coordinator remains the communication point through delivery, with the analysis, deliverables and timeline as agreed.
RESEARCH RESPONSIBILITY
Statistical Support Within Your Own Outcomes Study
OnlineSPSS handles the statistical design and the analysis of the comparison, while the clinical reasoning and its meaning for practice remain with your team.
Agreed Statistical Support OnlineSPSS provides the agreed methodological, analytical and statistical reporting support, defined in a scope that is confirmed before any work begins.
No Intellectual Contribution OnlineSPSS does not claim any intellectual contribution to your research. Authorship of the study report or article remains yours.
Your Research Decisions The research question, the clinical interpretation, the discussion and the conclusions remain yours, with overall responsibility.
Confidential Handling Research data and project details are handled confidentially, project information is not published, and an NDA is available where required.
FREQUENTLY ASKED QUESTIONS
Questions About Outcomes Research Analysis
Outcomes researchers often ask whether routine data can match a trial, why an active comparator helps and what to do when the database lacks a key clinical variable.
What does statistical analysis for outcomes research cover?
Statistical analysis for outcomes research covers studies that compare treatments or programs in routine practice: the cohort and time zero, control of confounding, the outcome models and the sensitivity analyses. Work can start at the protocol stage or after a reviewer’s report, and the clinical interpretation stays with you.
Can real-world data give the same answer as a randomized trial?
Sometimes the two agree, but not by default. Randomization balances unmeasured factors, whereas a real-world analysis can adjust only for what was recorded. A design that emulates a target trial, an active comparator and a bias analysis make the answer more credible, and the result is still described as an adjusted association unless its assumptions can be defended.
Why is an active comparator usually preferred to no treatment?
Patients who receive no treatment differ from treated patients in ways that are hard to measure, such as frailty or access to care. Two treatments used for the same condition are given to more similar patients, which reduces confounding by indication, though it does not remove it.
What if the database lacks a key clinical variable?
Claims data often lack laboratory values or disease severity. Proxies can be built from related codes, negative control outcomes can reveal residual bias, and a quantitative bias analysis shows how large the missing confounding would have to be. The limitation is stated in the interpretation and is not hidden.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as a protocol, a data dictionary or a database extract, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Start With Your Comparison
and Data Source
Describe the treatments or programs to be compared and the database behind them, with any reviewer or payer comments, and you will receive a proposed scope to approve before work begins.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014