From the clinical registry to a result a reviewer can verify
Survival, longitudinal data, predictive models and instrument validation: the statistical work that underpins a clinical study from start to finish.
Data from clinical practice are not necessarily collected to answer a research question. A registry can build up years of follow-up and still contain changes in criteria, loss to follow-up or events that do not lend themselves to a simple reading. A scale can produce a precise score and still fail to measure the construct of interest reliably. A predictive model can discriminate well and fall short exactly where it matters: in calibration, generalisation or clinical interpretation.
We work with clinical research groups, hospital departments and quality units. Knowledge of the care context and the clinical decision stay with the team that runs the study; our job is to turn the question into an analysis that respects how the data were produced, to estimate what the design allows to be estimated, and to quantify the uncertainty without hiding the limitations.
It is one of the areas where Azahar has the longest track record, with six published case studies and peer-reviewed research output on device survival, explainable predictive models and psychometric validation of clinical instruments.
We regularly work with
Hospital clinical research groups who want to take a question from the protocol to a methods and results section that can be reviewed and reproduced.
Departments and units with their own registries, built up over years, who need to know which questions that follow-up can really answer before choosing the model.
Teams who have developed, adapted or are using a questionnaire and need to show what it measures, how reliably and how its score should be interpreted.
Projects with small samples or difficult designs, where the point is not to force the usual analysis but to match the inference to the information that is actually available.
Where we usually come into a clinical research project
In clinical research, the entry point depends less on the name of the method than on the problem the study poses. Sometimes the decision is how to measure before collecting anything; at other times it is understanding a follow-up, or modelling risk or trajectories with the data already available; and in other projects the question lies in comparing performance, costs or patient flows. If you would rather start from what you need to do than from your sector, each block leads to the full path for that need.
Design a study
A study is won or lost before the first data point is collected.
When the sample size is too small for a frequentist analysis: prior knowledge built into the model and direct probabilities about the clinical hypothesis
View service →Validate the instrument and analyse what it measures: reliability, clinical scales and patient-reported outcomes
View service →I have data and don't know where to start
You already have the data. The question is what to do with them.
Time-to-event analysis for clinical prognosis
View service →From repeated measurements over time to an understanding of how things change
View service →Anticipating deterioration for early intervention
View service →Clinical segmentation for personalised medicine
View service →How patients move between states, where their trajectory changes and where follow-up is lost
View service →Evidence of effectiveness in routine clinical practice
View service →Find out what works
Did what you did work? And what is worth changing?
Resource optimisation without compromising quality
View service →Objective measurement of clinical performance
View service →Operational analysis to reduce overcrowding
View service →The rest of the catalogue, organised by what you need to do
Many of the problems in a health study cut across clinical areas. A database may need preparing before any model is fitted; an association may call for a better-defined causal question; a result may need visualisation, reproducible reporting or an independent second review. Those services are organised by the need they address, not by the hospital department the data come from.
Projects in this sector
What we are usually asked in this sector
We have a registry with years of follow-up. What can we get out of it?
It may hold a great deal of information, but the years accumulated do not on their own determine what can be concluded. First you need to know how events were defined, what happens when a patient stops appearing, whether the recording criteria changed and which variables were measured comparably throughout the period. From there, the decision is whether the question calls for survival analysis, competing risks, longitudinal models, transitions between states or another approach. The method comes after characterising the process that generated the data, because what each estimate really means depends on that process.
Our study has few patients. Is it worth analysing?
It can be, but a small sample means scaling back the ambition of the inference. No method can create information that the study does not contain. When there is sufficiently documented prior knowledge, a Bayesian approach can be useful for incorporating it explicitly and expressing uncertainty in directly interpretable terms; in other cases, the best decision may be to simplify the model, prioritise estimation over hypothesis testing or treat the work as exploratory. The choice depends on the question, the design and how much external support exists, not only on the number of patients.
We want to validate a scale we have developed. What does that involve?
It involves showing that the score the instrument produces can be interpreted as intended. Depending on the design and the stage of development, that may require studying its internal structure, consistency, stability over time and relationship with other measures it should converge with or diverge from. You also need to set how it is scored, how missing items are handled and which dimensions can be summarised without losing meaning. The result is not simply a Cronbach's alpha, but a documented justification of what the scale measures, how reliably and within what limits it can be used.
Does your work fit better elsewhere?
Shall we talk?
Tell us where your study stands and we'll give you a straight answer on how we can help.