Outcomes Model: Frequently Asked Questions
Is there a quick way of clearing data from the Input or Risk factor sheets?
11-May-2015
To clear all data from the Inputs sheet:
- Click on the Inputs tab
- Click in cell A17
- Hold down the ‘shift’ and ‘control’ keys, then press the ‘right arrow’ key
- Hold down the ‘shift and ‘control’ keys, then press the ‘down arrow’ key
- Press the ‘delete’ key
To clear all data from the Risk factor sheets:
- Click on the Smoking Status tab
- Shift click on the eGFR tab (this will select all sheets in between)
- Click in cell A17
- Hold down the ‘shift’ and ‘control’ keys, then press the ‘right arrow’ key
- Hold down the ‘shift and ‘control’ keys, then press the ‘down arrow’ key
- Press the ‘delete’ key
- Click on the Model parameters tab to deselect all other tabs
How do I optimise the number of processes to use?
11-May-2015
Detailed instructions for selecting the number processes to use for a given computer are provided in the OM2 manual.
What are the system requirements for Outcomes Model 2.1?
29-Jan-2020
OM2.1 has been formally validated on the following platforms:
Windows
- Windows 10 (64bit) and Microsoft Excel 2016 (64bit)
- Windows 7 (32/64bit) and Microsoft Excel 2013 (32/64bit)
Macintosh
- Mac OS 10.13 High Sierra and Microsoft Excel 2016
- Mac OS 10.14 Mojave and Microsoft Excel 2016
- Mac OS 10.15 Catalina and Microsoft Excel 2016
- Informal testing with Excel 2011
Other combinations have been checked informally and shown to work but we would suggest different combinations be tested against one of those listed above.
WHAT IF I HAVE NO DATA ON RISK FACTORS POST-BASELINE?
26-Oct-2023
The UKPDS Outcomes model allows entry of baseline risk factors and also annual risk factor values for each patient. If you do not have real or simulated data for the annual values, there are simple options provided to populate the sheet automatically and these are explained in the user manual (Risk Factor worksheets).
In version 2.2 a new option was introduced using built in risk factor trajectories based on a paper by Leal et al, which uses data from the UKPDS trial to extrapolate risk factors like HDL, BMI and smoking status from baseline onwards and populate the annual risk factor tabs (DOI: 10.1111/dme.14656).
To use these equations enter Method 3 (= Use UKPDS prediction formula) in individual risk factor sheets and enter "Y" in the "Risk factor equations" box (N14) in the Model Parameters sheet.
For HDL, LDL, Systolic BP, HbA1c, Weight, Heart rate, WBC, Haemoglobin these are calculated by Excel Macros using the "Populate annual risk factor sheets" button before running the model.
For Smoking Status, PVD, AF, Albuminuria, eGFR the equations incorporate uncertainty and are calculated during model execution. Leave the values blank to allow this.
If you have post-baseline data on some risk factors, but not all (or have missing follow up data for only some patients), you can enter the data you do have and leave the others blank, making sure that you enter “N” in the “Replace existing :” box on the risk factor sheets (B6), so that you preserve the values you enter on each sheet and just fill in the gaps.
What do I do if I have individual patient data on some but not all baseline risk factors?
26-Oct-2023
Other researchers have used different approaches to solve this problem. Pagano et al (DOI: 10.1111/dom.14311) used UKPDS trial data to estimate regression equations that can be used to impute values for heart rate, haemoglobin, white blood cell count and history of PVD, atrial fibrillation, CHF, IHD, renal failure, amputation, blindness and foot ulcer. The coefficients are on pages 9-10 of the supplementary material. The paper describes how to convert the logistic regression coefficients into unbiased binary data showing whether or not patients had each event that replicate the proportion of patients in UKPDS that had each event.
Researchers using UKPDS-OM2 before publication of the Pagano papers have also applied a population mean for risk factors that are not recorded in their sample; such population means may come from previous studies, such as the Lipids in Diabetes study (DOI: 10.1007/s00125-013-2940-y) or UKPDS. This approach could be used if you have insufficient data to apply the Pagano equations, although it will underestimate heterogeneity between patients and correlations between risk factors, thereby potentially leading to underestimation of uncertainty and incorrect estimates: especially if it is applied to strong model drivers, such as LDL or blood pressure.
Randomised trial data, cohort studies or clinical audits may be good sources of individual patient data on risk factors. The most important thing is to get a sample of patients that is representative of the population of interest. If you plan to apply treatment effects from a randomised trial or meta-analysis to a single sample of patients, a small but representative sample of patients may be sufficient.
If you have no data on baseline risk factors, data could in principle be simulated from secondary data on the mean and standard deviation for another population, although simulated data must be within the range of values allowed in UKPDS (see supplementary table 3 of Pagano et al DOI: 10.1111/dom.14311) and should ideally take account of the distribution of risk factors and correlations between risk factors (e.g. patients with high BMI may also be more likely to have high LDL and mean values of many risk factors differ between men and women).