Analytics Foundations: From Data to Action
Analytics is not charts for their own sake — it is the disciplined path that turns routine health data into evidence for saving mothers’ lives. This session lays the conceptual groundwork for the Superset lab that follows: the data-to-action chain, the four types of analytics, the national data sources, how data quality is scored before any metric is trusted, and the analytical questions that every chart exists to answer.
- Explain analytics as a chain that turns raw data into information, insight, and action
- Distinguish descriptive, diagnostic, predictive, and prescriptive analytics using maternal-health examples
- Compare the roles of DHIS2, MPCDSR, and DHS as complementary data sources
- Score a data source against the six data-quality pillars and compute its weighted overall grade
- Map the four analytical questions that the Superset lab's dashboard tabs are built to answer
From raw records to action
Analytics is a chain. It moves from raw records, to information, to insight, to action — and each link adds meaning that the previous one lacks.
Consider a single maternal death record. As data, it is one row among many: a birth, a death, and a cause recorded in DHIS2 or MPCDSR. Aggregated and computed into information, it contributes to the maternal mortality ratio and to distributions that reveal where deaths cluster, why they happen, and to whom. Interpreted as insight, those patterns become evidence-based recommendations — what to do and where to focus first. And as action, the recommendation becomes a targeted intervention, a resourcing decision, or a policy change. Each maternal death record, handled this way, is a step on the way to an intervention that saves the next mother.
The four types of analytics
The four types build on each other, rising in value and complexity as they move from describing the past to prescribing the future.
| Type | Question it answers | Maternal-health example |
|---|---|---|
| Descriptive | What happened? | MMR is 77 per 100,000; 445 deaths from obstetric haemorrhage |
| Diagnostic | Why did it happen? | 81% of deaths were potentially avoidable; care quality is the leading delay |
| Predictive | What will happen? | Forecast the MMR trajectory against the 86 target for 2025/26 |
| Prescriptive | What should we do? | Prioritise PPH management and strengthen referral pathways |
The programme starts with descriptive and diagnostic analytics — establishing reliably what happened and why — and builds toward predictive and prescriptive work from that foundation.
Data sources
Three complementary sources feed the maternal mortality analytics, each answering a different part of the picture.
| Source | Role | What it provides |
|---|---|---|
| DHIS2 | Routine | Births, maternal deaths, and causes of death. Monthly reporting at national coverage — the numerator and denominator for the MMR |
| MPCDSR | Surveillance | Maternal and perinatal death surveillance and response. Case-level audit detail: the three delays, referral status, avoidability, and risk |
| DHS | Population | The Demographic and Health Survey. A periodic population-based MMR estimate used to validate the routine figures |
DHIS2 gives the routine monthly numbers; MPCDSR explains why each individual death happened; DHS provides the independent benchmark the routine estimate is checked against.
Assessing data quality
A dataset is only as good as the quality behind it. Before any metric is trusted, every source is scored against six quality pillars. Each pillar is rated on a 1–5 scale, and the pillars are weighted — completeness and consistency carry the most weight in the overall score.
| Pillar | Definition | Weight |
|---|---|---|
| Completeness | Are all required data fields filled in? | 40% |
| Consistency | Is the data consistent across systems and within the dataset? | 30% |
| Accuracy | Is the data correct and free from errors? | 10% |
| Timeliness | Is the data up-to-date and relevant at the time of analysis? | 10% |
| Validity | Does the data conform to the required format and business rules? | 5% |
| Uniqueness | Are there duplicate records? | 5% |
On the 1–5 pillar scale, 1 is unacceptable, 2 poor, 3 adequate, 4 good, and 5 excellent. The weighted pillar scores roll up to an overall grade:
| Overall score | Grade |
|---|---|
| 80–85% | Acceptable |
| 85–90% | Good |
| 90–95% | Very Good |
| 95% and above | Amazing |
The assessment is applied across the routine sources — DHIS2, IDSR, and HMIS — so their quality can be compared side by side, and weaknesses addressed before the data drives a decision.
The four analytical questions
Four questions drive every chart built in the lab. Keep them in mind — the dashboard tabs you will assemble map directly onto them.
- Overview — what is the trend, geographical, and hospital distribution of maternal mortality?
- Causes — how do direct and indirect deaths break down: haemorrhage, sepsis, hypertensive disorders?
- Associated factors — what role do referral, high-risk pregnancy, avoidability, age, and parity play?
- Comparison — how does the routine (DHIS2) MMR compare against the DHS survey estimate?
A useful reflection before moving on: which of these four questions can your national dashboards answer today — and which data gap blocks the rest? With the foundations in place, the next session builds the answers in Superset.