RESEARCH DISCIPLINE
Biostatistics for Health Data Research
Built for Care, Analyzed for Research
Electronic health record data analysis for studies that also draw on claims and registries: codes turned into defined cohorts, the reasons for absent values understood, and confounding addressed before any estimate is reported to a reviewer.
No Commitment Required
Strictly Confidential
In-House PhD-Level Statisticians
Fixed Project Fee
Same-Day Response
What Does Health Data Biostatistics Involve?
Health data biostatistics applies biostatistical methods to records created for care or billing, where the research question arrived after the data. EHR data analysis, like work with claims and registry extracts, has to turn diagnosis, procedure and medication codes into defined exposures and outcomes, explain why values are absent, and adjust for confounding as far as the design allows. OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, supports clinical investigators and graduate researchers working with routinely collected health data: an in-house PhD-level statistician builds the analysis on your extract and keeps the limits of the source system in view throughout. What the findings mean for patients and practice is for you to decide.
WHEN TO ENGAGE
Before, During and After the Extract
The questions differ before the extract is pulled, once the file is in hand, while a model is chosen and after a reviewer has queried the cohort.
Before the Extract The study period, the codes that define the cohort and the variables to be pulled are fixed with the analysis in mind, so a field needed later is not found missing after extraction.
File Received A file with one row per encounter, visit or laboratory result has arrived, and it must be reshaped to one row per patient, with baseline, exposure and follow-up defined, before a model is chosen.
Gaps in the Records Laboratory values, vital signs or follow-up visits exist for only some patients, and the reasons they are absent must be understood before any rule for handling them is applied.
Validity Questioned A reviewer or committee has queried how a condition was identified from codes, who was lost to follow-up or whether the comparison is confounded, and the answer needs evidence from the records.
WHAT THE SUPPORT COVERS
From Code Lists to Reported Estimates
Most of the statistical effort in a record study goes into decisions made before any model is fitted, and the support covers those decisions as fully as the models.
Code-Based Definitions
Diagnosis, procedure and medication codes become exposure, outcome and covariate definitions, with the look-back window stated and the accuracy of the codes checked against chart review where that is available.
Cohort Entry and Follow-Up
Entry, baseline and the start of follow-up are fixed for every patient, because the first record in a system is often not the start of the disease or of the treatment being studied.
Absent Values
Absent laboratory results are examined through Missing Data Analysis, since tests are ordered when a clinician is concerned, so a missing value says something about the patient.
Confounding Control
Adjustment sets are chosen from clinical reasoning about treatment decisions, not from the data alone, in line with the confounding principles of Epidemiology.
Linked and Repeated Records
Patients with many encounters, or seen in several clinics, are linked and counted once, so that frequent users do not appear as many independent people and one person is not split into several.
Reporting Record Studies
Findings follow the RECORD extension of STROBE, with the code lists supplied, as expected in Biostatistics and Health Sciences research.
COMMON ISSUES
Weaknesses Built Into Record-Based Studies
Because the records were created for care, these weaknesses are present in the file from the start, and a reviewer of a record-based study will look for each one.
Absence Read as Normal A laboratory test with no recorded result is coded as normal or as zero, though clinicians order tests selectively, and the comparison between groups is biased in a direction that is hard to predict.
Codes Taken at Face Value A diagnosis code is treated as proof of disease, though billing codes can miss true cases and include false ones, and their accuracy often differs between hospitals and across years.
Sicker Patients Seen More Patients with more visits have more chances of a diagnosis and more measurements, so counts of tests and diagnoses rise with illness and access to care, not only with the exposure under study.
Lost Follow-Up Mistaken for Survival A patient who stops appearing in one health system is treated as alive and event-free, though they may have moved to another system, which can bias outcome rates downward.
Reason for Treatment Ignored Treated and untreated patients differ because of why treatment was chosen, so adjusting for age and sex alone leaves the main source of bias in the comparison untouched.
One Patient, Several Identities Merging records from clinics without a reliable identifier splits one patient into several or joins two into one, which alters the size of the cohort and the number of events.
DELIVERABLES
Documents That Travel With Your Results
The definitions and exclusions behind each estimate are recorded in the files below, so that another analyst could rebuild the cohort from the same source.
Code Lists and Definitions Every diagnosis, procedure, medication and laboratory code used to define the cohort, exposure, outcome and covariates, with the look-back window and the source of each list.
Cohort Flow Record A step-by-step account of how many patients each inclusion and exclusion rule removed, from the raw extract to the analytic sample, with the reason for each rule.
Record Completeness Summary A summary by variable of how often values are recorded, in which patient groups and clinics, which guides how absent values are handled in the models.
Adjusted Estimate Tables Adjusted estimates with 95% confidence intervals and the number of patients and events behind each, laid out for a manuscript or a thesis.
Robustness Checks Results under alternative code definitions, look-back windows and handling of absent values, showing which findings hold across reasonable choices.
Source-Data Caveats A short note on what the source system does not record, such as care received elsewhere, listed as a limitation of the data, with the scripts attached.
ILLUSTRATIVE EXAMPLE
Blood Pressure Records in Outpatient Clinics
An illustration of how a question about control of a chronic condition is analyzed when data exist only for patients who attended.
ILLUSTRATIVE EXAMPLE · OUTPATIENT RECORD DATABASE RESEARCH · UNITED STATES
A graduate researcher in clinical informatics wanted to describe blood pressure control among adults with hypertension in the electronic records of several outpatient clinics, and to compare patients whose medication was changed with those whose medication was not.
The cohort was defined from diagnosis codes and visits before any outcome was examined, with a look-back window and a rule for the minimum number of readings. The complication lay in the readings themselves: patients with poorer control visited more often, so they contributed more measurements, while patients with no reading in the year could not be classed as controlled or uncontrolled. The pattern of recording was described by clinic and patient group, the primary analysis used one value per patient in a fixed window, and a second analysis handled the absent readings under a stated assumption. The comparison was reported as an adjusted association, with the reason for medication changes discussed as a limit.
Outcome: The researcher received the adjusted difference in control between the groups with its 95% confidence interval, the record completeness summary, the cohort flow record, the code lists and the scripts.
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
Matching the Tool to the Extract
The tool follows the format of the extract and the habits of your group, and the code is delivered so that another member of the team can rerun the analysis.
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
What Remains With You in a Health Data Study
OnlineSPSS handles the statistical treatment of the records, while the clinical question and what the findings mean for care stay with the research 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 article or thesis built on these records remains yours.
Your Research Decisions The research question, the clinical interpretation of the findings, 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 Analyzing Routinely Collected Health Data
Records collected for care leave gaps that any reader of the study will probe, and the first answers here start from those gaps.
What does biostatistical analysis of health records cover?
The analysis of electronic health records covers the statistical design and analysis of studies that use records created for care or administration: code-based definitions, cohort entry and follow-up, absent values, confounding control and reporting. It applies equally to claims and registries, and the clinical meaning of the findings is yours to judge.
How accurate are diagnosis codes for identifying a condition?
Accuracy varies by condition, setting and coding practice. Some codes identify a disease well, while others miss cases or include patients who do not have it. A definition can be compared with chart review in a sample, alternative definitions can be run as a robustness check, and the choice is reported so that a reader can judge how far misclassification could move the estimate.
Is a missing laboratory value the same as a normal one?
No. Tests are usually ordered for a reason, so an absent result often means the test was not thought necessary or was done elsewhere. Treating it as normal, or dropping those patients, can bias a comparison. The pattern of recording is described first, and the handling, for example multiple imputation under a stated assumption, is chosen to match it and checked against the complete-patient result.
What if my records stop when a patient leaves the hospital?
Outcomes that occur elsewhere are invisible in one system’s records, so follow-up looks shorter and events look rarer than they were. Follow-up ends at a patient’s last contact, linkage to death registries or other sources is used where it is available, and the limitation is stated with the results.
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, the code lists or a data dictionary for the 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
Records Collected for Care,
Now a Study?
Every estimate drawn from health records inherits the code lists behind it, so those definitions are where the statistical work starts.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014