# Answer key · MG10

## What share of heart-failure patients with and without diabetes had a recorded death?

Diabetes is one of the conditions recorded for every patient in this heart-failure cohort, and clinicians want to know whether it marks patients at higher risk. A crude death count is the first thing people look at, so it has to be read with its limits in view.

- One record is — one patient record
- Period — April–December 2015
- Scope — Patients with heart failure at two hospitals in Faisalabad, Pakistan.

## The table

Follow-up outcome × Diabetes

| Follow-up outcome | Diabetes | No diabetes |
| --- | ---: | ---: |
| Death recorded | 40 | 56 |
| No death recorded | 85 | 118 |

## What can this table say about diabetes and recorded deaths in this cohort?

1. In this cohort, the share with a recorded death was almost the same with and without diabetes.
2. Diabetes did not change these patients’ risk of death.
3. Each share is the group’s mortality at a fixed time.

Choose one statement. Show the counts and denominators that justify it, then describe one limit of the comparison.

## The figures behind the check

- **1. Supported.** 40 of 125 (32.0%) against 56 of 174 (32.2%), a gap of 0.2 percentage points. It describes the 299 released records from two hospitals, nothing beyond them.
- **2. Not supported.** Follow-up time varied between patients, and a record without a death may simply have been followed for less time. Equal crude shares do not show equal risk; comparing risk needs a survival analysis that uses follow-up time.
- **3. Not supported.** No fixed time point is used. Each share counts deaths recorded during whatever follow-up each patient had, so it is a crude event proportion.

- Diabetes: 40 of 125 with a death recorded (32.0%).
- No diabetes: 56 of 174 with a death recorded (32.2%).
- Whole cohort: 96 of 299 with a death recorded; follow-up length differs by patient.

Quote this as a description of one historical cohort: crude shares, unequal follow-up, no adjustment. It is a methods example, not guidance on diagnosis or treatment.

## Limits

A selected clinical cohort with follow-up durations that vary from patient to patient. A crude death-event proportion does not give mortality at a fixed time or the effect of any condition or treatment. This is not medical advice.

## Source

Data: Heart Failure Clinical Records (2020), UCI Machine Learning Repository. Licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Crosstabs selected, recoded and aggregated the records into this table.

- [Heart Failure Clinical Records (2020), UCI Machine Learning Repository.](https://archive.ics.uci.edu/dataset/519)
- [Reproduce it · MG10](https://www.crosstabs.com/examples/mg10-recorded-deaths-by-diabetes-status)
- [CSV](https://www.crosstabs.com/examples/mg10-recorded-deaths-by-diabetes-status/download/csv) · [JSON](https://www.crosstabs.com/examples/mg10-recorded-deaths-by-diabetes-status/download/json)
