- Home
- /
- By your needs
- /
- I have data and don't know where to start
You already have the data. The question is what to do with them.
Having data is not the same as having answers. Between a database and a defensible conclusion lies a piece of analytical work that depends on which questions you want to answer and what kind of data you have in front of you.
Arriving here with data already collected is the most common situation: a registry that has kept growing, the results of a study, a clinical database, a cohort that has been running for years. The material is there, but turning it into useful knowledge means choosing the right method, and that choice depends on the question and the nature of the data, not on whichever technique happens to be in fashion.
Analysing data can mean very different things depending on the question you want to answer: studying how an outcome evolves, explaining relationships and mechanisms, predicting a risk or segmenting a population, working with omics data, or synthesising the evidence that already exists. You don't need to know which method you need: it is enough to know what you want to find out.
Start with the strand you recognise as yours. If your question is not yet clear, the first step is to narrow it down from your objectives and the information available; if you already know what you want to find out, go straight to the relevant section.
This is for you if…
You have data already collected (a registry, a cohort, the results of a study) and need a rigorous statistical analysis, but you are not sure which method is the right one.
You know what you want to find out (whether a treatment is associated with better survival, which patients are at higher risk, what explains an outcome) but not how to set up the analysis.
You have complex data (longitudinal, survival, omics, high-dimensional) that cannot be analysed with basic statistical tools.
You need to synthesise the evidence that already exists on a question, or a methodological second opinion on an analysis you already have.
How we can help
Analyse outcomes and how they evolve
Study how an outcome behaves and how it evolves over time.
The general statistical analysis of a dataset: hypothesis tests, regression models, estimation of associations, with the choice of method guided by the nature of the data and the question. The starting point when what is needed is a sound, well-grounded analysis, without a specialised method involved.
View service →When what is measured is the time until an event occurs (a relapse, a discharge, a failure), and not every subject experiences it during follow-up. Kaplan-Meier, Cox models, competing risks: the family of methods that handles censored data correctly instead of ignoring them.
View service →When the same subjects are measured repeatedly over time and those measurements are not independent of one another. Mixed-effects and random-effects models that account for within-subject correlation, to study how a variable evolves without treating each measurement as if it came from a different person.
View service →Generating evidence on outcomes, utilisation and effectiveness in routine practice from registries, clinical records or observational cohorts, rather than from the controlled conditions of a trial. Useful when you want to know what happens in real-world practice, with the methodological rigour that non-experimental data require.
View service →Explain relationships and mechanisms
Study causal relationships and possible mechanisms under explicit assumptions.
When the question is whether A caused B and not just whether they go together, causal inference provides the framework for estimating effects from observational data: control of confounders, matching, weighting, instrumental variables. Estimating a causal effect outside a randomised trial, with the assumptions always on the table.
View service →When you want to study through which mechanisms an effect may operate and whether it varies across contexts or subgroups. Estimation of direct and indirect effects and interaction analysis, with the causal and measurement assumptions made explicit.
View service →Predict, segment and study trajectories
Anticipate an outcome, group subjects or follow their path.
Building models that estimate the probability of an outcome from each subject's characteristics: regression, machine learning, performance validation. Predictive models assessed honestly (calibration and discrimination, not just a single accuracy figure) and with the interpretability that the clinical context demands.
View service →Grouping patients into profiles with similar behaviour or risk, to understand the heterogeneity of a population instead of treating it as a single block. Clustering and segmentation methods applied with clinical judgement, not just statistical criteria, so that the resulting groups mean something.
View service →Reconstructing the path patients take through a care process: which states they pass through, in what order, how long they remain in each one. Trajectory models and state-transition chains to understand how a patient population moves through the system, beyond a static snapshot.
View service →When what matters is how an indicator evolves over time: decomposing a series into trend, seasonality and irregular component, detecting change points, modelling temporal dependence to assess whether the observed change exceeds the expected temporal variation. The tool for reading how a quantity changes over time.
View service →When the object of study is the structure of relationships (between patients, between professionals, between variables), network analysis makes it explicit: who is connected to whom, which nodes are central, how they cluster. A lens for reading the structure that underlies a system, beyond counting isolated elements.
View service →Analyse biomarkers and omics data
Study biological signals, from marker panels to high-dimensional omics data.
Studying the relationship between biological markers and a clinical outcome when there are multiple correlated markers: selection of relevant variables, control of multiplicity and models that separate signal from noise. The statistical analysis that turns a panel of markers into interpretable information.
View service →The analysis of genomic, transcriptomic and other omics data, where the number of variables far exceeds the number of samples. Specific high-dimensional methods, correction for multiple comparisons and control of overfitting, applied with the rigour required in a field where spurious patterns are easy to find.
View service →Synthesise, review or map evidence
Integrate study results, review an existing analysis or map the scientific output of a field.
Quantitatively combining the results of several studies on the same question to estimate a pooled effect and examine its consistency: random-effects models, heterogeneity diagnostics, assessment of publication bias. Integrating scattered evidence into a well-founded estimate, with its limits made explicit.
View service →An independent review of a study's statistical approach: whether the chosen method is appropriate, whether the assumptions hold, whether the conclusions follow from the data. An external view of an analysis, your own or someone else's, before publishing, responding to a reviewer or making a decision based on its results.
View service →The quantitative analysis of the scientific output on a topic: what is published, by whom, how it evolves, how it connects. Mapping the output, evolution and structure of a research field with reproducible methods, useful as the basis for a review or a research strategy.
View service →Not sure which of these fits your situation?
Tell us what stage you are at and we will tell you what work is needed, or whether none is. No commitment, and to the point.
Let's talk about your dataHow we have approached it in other projects
Before you take the step
Do I need to know which statistical analysis I want before getting in touch?
No: arriving without knowing which method you need is the normal situation, not an exception. This is for anyone who has data already collected (a registry, a cohort, the results of a study, a clinical database) and needs a rigorous analysis, but is not sure which one is right. Choosing the method is part of the work, and that choice depends on the question you want to answer and the nature of the data, not on whichever technique happens to be in fashion. It is enough to know what you want to find out: whether a treatment is associated with better survival, which patients are at higher risk, what explains an outcome. From there, it is decided whether the problem calls for survival analysis, a predictive model, causal inference or another approach.
I have several objectives in the same study. How do you decide which analysis each one needs?
It is decided by reading each objective separately and seeing what type of question it poses, because different objectives usually require different methods on the same data. This is common when a single database has to answer several things: describing how an outcome evolves, explaining a relationship, predicting a risk, segmenting the population. A single project can combine a descriptive, a causal and a predictive analysis without any inconsistency: they are different questions, each answered with the right tool. What we avoid is applying a single method to everything out of inertia. The analysis plan assigns each objective the approach that answers it, instead of forcing one technique onto questions that are not the same.
What information do you need to propose an analysis strategy?
We need to understand what data you have and what you want to answer with them: the structure of the database (which variables, how many subjects, how they were measured), the outcome of interest and the question or questions of the study. The database does not need to arrive clean, and you don't need to know which method to apply; it does help to know how the data were collected and whether there are repeated measures, censoring, many variables or few subjects, because those characteristics shape the method. With that, we assess whether the data can answer the question as it stands, or whether it is better to reformulate it or strengthen the database first. Sometimes the initial review reveals that the data do not support the original question, and saying so in time is part of the work.
My data are complex (longitudinal, survival, omics, high-dimensional). Do they belong here, or do I need something special?
They belong here fully: this covers both general analysis and the specialised methods that complex data require. This applies when the database cannot be analysed with basic statistical tools: repeated measures on the same subjects, time to an event with censoring, biomarker panels, omics data with more variables than samples. Each of those structures has its own family of methods. You don't need to identify which case is yours; recognising that your data are complex and do not fit a standard analysis is enough. What we don't do is treat complex data with a method that ignores their structure (analysing repeated measures as if they were independent, for example): the analysis respects the real nature of the data.
What happens if, when the data are reviewed, it turns out they cannot properly answer my question?
If the data do not support the original question, we say so clearly and set out the realistic options, instead of forcing a result that does not hold up. It happens when the review reveals that key variables are missing, that the sample is too small for the effect sought, that there is a bias that shapes the interpretation or that the outcome is not well measured. The honest options are usually three: reformulate the question towards what the data can answer, strengthen the database where feasible (a preliminary data quality phase, for example), or narrow the conclusions while making the limitations explicit. What we don't do is dress up an analysis so that it appears to answer something it cannot: an impeccable analysis on data that cannot support the question produces a convincing and wrong result.
I already have an analysis, my own or another team's. Can you review it instead of redoing it?
Yes: reviewing an existing analysis is a starting point in its own right, distinct from analysing from scratch. It fits when you want an independent view before publishing, responding to a reviewer or making a decision based on the results: checking whether the chosen method is appropriate, whether the assumptions hold and whether the conclusions follow from the data. It does not always mean redoing the work; often it is enough to review the approach and point out where it can be strengthened. When the review finds that the analysis needs to be partly redone, we say so and scope that work separately. That is what the methodological second opinion covers: an external judgement on the soundness of an analysis, before its conclusions are made public or turned into a decision.
An analysis produces results that almost always need to be communicated, documented or turned into a tool that others can use. When you have the result and need to put it to work, the natural next step is to communicate results and build tools that decision-makers can work with. And if you are still looking at your data without knowing what to ask, let's talk: sometimes the first step is to narrow down the question.
Shall we talk?
Tell us what you are working on and we will give you a straight answer on how we can help.