How To: Analyse & Present Data

INTRODUCTION

The aim of this ‘How To’ guide is to provide advice on how to analyse your data and how to present it. If you
require any help with your data analysis please discuss with your divisional Clinical Audit Facilitator who will
be happy to help.

1. HOW TO ANALYSE DATA

CALCULATING COMPLIANCE WITH CLINICAL
Audit data comes in three different forms, ‘tick-box’, AUDIT STANDARD
numerical or freetext. Each requires different methods
of analysis, but in each case the aim is to establish which Number of patients
standards are being met (% compliance) and which are who meet standard
not (% non-compliance). If a standard is not being met 100
you need to identify why and how practice can be x
improved to ensure that the standard is met in the Number of Number of
future. You may also consider if there were other, patients to patients who
acceptable reasons for the standard not being met, i.e. whom standard
– meet any listed
an exception not considered during the planning stage. applies exceptions

A. TICK-BOX DATA

It is likely that the majority of the data that you have obtained from your data collection form will relate to
yes/no options or tick-box options from a specified list of alternatives. In such cases, it is usual practice to
add up the number of answers recorded for each option and express the total as a raw number and as a
percentage.

EXAMPLE 1:
• Sample size: 50 patients
• Audit criteria: All patients should attend a pre-operative clinic
• Question: Did the patient attend a pre-operative clinic?
• Results: Yes = 32 and No = 18.

A good way of expressing this data is:

• All patients should attend a pre-operative clinic. n=50

• Yes = 32 (64%)
• No = 18 (36%)

The ‘n=50’ indicates how many patients were in the audit sample and is used to calculate the
percentages, i.e. 32/50 = 64%.

It is important to remember that yes/no options do not allow for ‘not applicable’ answers. Taking the
example used above, it is possible that certain patients did not meet the standard because they had an
emergency operation. In this instance the answer to the question ‘Did the patient attend a pre-operative
clinic?’ would have been ‘not applicable’. To reflect this the data can be expressed more accurately as:

 2009 UHBristol Clinical Audit Team – Version 3 Page 1 of 8

 How To: Analyse & Present Data

EXAMPLE 2:
• Audit criteria: All patients should attend a pre-op clinic 32 32
• Exception: emergency operation 100
= x
• Results: Yes = 32, No = 13 and N/A (emergency) = 5
50 – 5 45

• 32 patients attended a pre-op clinic

• 18 did not, but 5 of these were emergencies (exception criteria)
• Therefore 32/45 (71%) met the standard

B. NUMERICAL DATA

Some of the data items you collect are likely to be numerical values, e.g. age, length of stay in hospital,
blood glucose level etc. Lists of numbers like this can be summarised using measures of central tendency
and dispersion:
• Measures of central tendency look at the middle/common values in a list of data items: the mean,
median and mode.
• Measures of dispersion look at how spread the data is: the range.

MEASURES OF CENTRAL TENDENCY

The mean is the average value, calculated as: Sum of all the values
Number of values

The table below shows data about length of stay (LOS) on three wards:

Number of patients discharged

Length of Stay (days)
Ward 1 Ward 2 Ward 3
1 4 4 1
2 8 7 3
3 12 17 3
4 18 10 4
5 20 7 10
6 18 4 15
7 12 2 4
8 8 2 2
9 3 5 0
10 0 9 0

For Ward 1, the mean is:

(1x4) + (2x8) + (3x12) + (4x18) + (5x20) + (6x18) + (7x12) + (8x8) + (9x3) = 511 = 4.96
4 + 8 + 12 + 18 + 20 + 18 + 12 + 8 + 3 103

The mean LOS on ward 1 is 5 days (rounded to nearest whole day).

If the same formula were used to calculate the means for wards 2 and 3, you will find that for each ward,
the mean LOS is 5 days. However, the mean is not always the best measure of central tendency.

The LOS for all three wards is illustrated on the graph below. The mean suggests that the data is the same
for all three wards, however the graph indicates that this is not the case. The problem with the mean is what
it does not tell us.

 2009 UHBristol Clinical Audit Team – Version 3 Page 2 of 8

 How To: Analyse & Present Data

Length of Stay

Number of patients discharged

25

20
Ward 1
15
Ward 2
10 Ward 3

0
1 2 3 4 5 6 7 8 9 10
Day

The data collected for Ward 1 is almost perfectly symmetrical, with the graph illustrating that the data
follows the shape of a ‘bell curve’. Data that conforms to this shape is known as ‘parametric’ data. In this
instance the mean is an appropriate measure of central tendency.

The data for Wards 2 and 3 is non-parametric. Their graphs do not form a symmetrical curve. Describing
their notable features, Ward 2 has a significant proportion of patients with a LOS of 3 days together with a
number of patients staying 9 or 10 days. Ward 3 has a peak LOS of 6 days. It can be seen that using the mean
alone with non-parametric data is not very informative. The median and mode can help to convey the
missing information.

The mode is the most commonly occurring value. For Ward 2 this is 3 days and for Ward 3 it is 6 days. This
should be obvious from both the raw data and the graph. If the highest occurrence is shared by more than
one value you could either state them all as modal values, or none. For example, if for Ward 3 there were
10 patients discharged on both day 5 and day 6 you could either say there were 2 modal values of 5 and 6,
or that there was no mode.

The median is the mid-point of all the values. For Ward 2, we have data on 67 patients. If we made a list of
LOS, placed in order from the lowest to the highest, the mid-point would be the 34th value i.e. there are 33
values below and above this. The 34th value relates to a patient who was discharged after 5 days, so this is
the median. For Ward 3, we have data on 42 patients, i.e. there is no single mid-point. In this case, take the
average of the 21st and 22nd value (there are 20 values below and above these two values). The 21st value
relates to a patient who was discharged after 5 days and the 22nd value relates to a patient who was
discharged after 6 days, so the median is 5.5 days (5+6 divided by 2).

Unless you are well versed in statistics, we would advise that you use all three measures of central tendency,
or show the information using a graph. In general, quote median rather than mean for non-parametric data.

Not all lists of numerical data should be analysed in this way. For example, if your standard is ‘The patient
will be considered medically fit for surgery if temperature <38°C’ and you collect a list of temperature data,
it would not be meaningful to present the mean, median and mode temperature. What you are interested
in here is the percentage of cases that met the standard i.e. the percentage of surgical cases with
temperature <38°C.

MEASURES OF DISPERSION
As well as stating the mean, median and mode, it is also good practice to provide some indication of how
spread the data is. The range states the lowest and highest values. In our example:

Ward 1 has a range of 1-9 days

Ward 2 has a range of 1-10 days
Ward 3 has a range of 1-8 days

 2009 UHBristol Clinical Audit Team – Version 3 Page 3 of 8

 How To: Analyse & Present Data

A more subtle way of expressing dispersion is to use quartile range. This involves listing your values from
lowest to highest, as per calculating the median, and then dividing the values into four equal parts or sub-
ranges. The range you are interested in lies between the second and third quarter (or ‘quartile’).

So, for example, some more LOS data:

Ward C: 1 2 3 5 | 5 6 6 7 | 7 7 7 9 | 9 11 20 38

In this case the range is 1-38 days, but the quartile range is 5-9 days

The quartile range is useful in taking out outlying data (data some distance away from the
mean/median/mode), as in the case of Ward C above and Ward 2 in our first example. Ward 2 has the
largest range but a comparable quartile range to Wards 1 and 3:

Ward 1 - range 1-9 days; quartile range 4-6 days

Ward 2 - range 1-10 days; quartile range 3-7 days
Ward 3 - range 1-8 days; quartile range 4-7 days

ANALYSING DATA AGAINST STANDARDS

If your standard statement was ‘Patients should be discharged by the end of their 5th day following surgery’,
using LOS data for ward 2, you find that 45 out of the 67 discharged patients had a LOS of 5 days or less.

You would write this as 45/67 (67%) patients met the standard.

C. FREE-TEXT DATA

If you include an open question in your data collection form, you will obtain free-text data. In order to
analyse this data you should group comments into themes or categories, i.e. as if you were creating tick-box
options for the data collection form. You might also want to consider reproducing some comments verbatim
in your report, if they are particularly pertinent.

DRAWING CONCLUSIONS

The end stage of your analysis is concluding how well the standards were met and, if applicable, identifying
reasons why the standard was not met in all cases. These reasons might be agreed to be acceptable, i.e.
could be added to the exception criteria for the standard in future, or will suggest a focus for improvement.
In theory, any case where the standard (criteria or exceptions) was not met in 100% of cases suggests a
potential for improvement in care. In practice, where standard results were close to 100%, it might be
agreed that any further improvement will be difficult to obtain and that other standards, with results further
away from 100%, are the priority targets for action. This decision will depend on the topic area, in some ‘life
or death’ type cases, it will be important to achieve 100%, in other areas a lower result might still be
considered acceptable.

2. DISPLAYING DATA

CONTINUOUS DATA

‘A set of data is said to be continuous if the values/observations belonging to it may take on any value within
a finite or infinite interval. You can count, order and measure continuous data. For example height, weight,
temperature, the amount of sugar in an orange, the time required to run a mile’.
http://www.stats.gla.ac.uk/steps/glossary/presenting_data.html

 2009 UHBristol Clinical Audit Team – Version 3 Page 4 of 8

 How To: Analyse & Present Data

Continuous data fits on a numerical scale. Examples of continuous data include:

• Temperature: 34°, 35°, 36°, 37°, 38°, etc.
• Days post-op: 1, 2, 3, 4, 5, 6, 7, 8, etc.
• Age: 16, 17, 18, 19, 20, 21, etc.

Generally when displaying continuous data, it would be appropriate to use charts that allow you to lay out
this numerical scale and plot data against it e.g. scatter graphs, box and whisker plots, etc.

CATEGORICAL DATA

Categorical data is data that can be sorted according to non-overlapping (mutually exclusive) categories,
whereby each subject in a sample can only fit into one category. For example:
• Gender: Male/ Female
• Age group: 16-20, 21-25, 26-30, etc.
• Standard met: Yes / No

The usual type of data resulting from a clinical audit is categorical data and would most commonly be
represented as a bar chart or a pie chart.

PIE CHARTS AND BAR GRAPHS

Bar or pie charts are most commonly used in clinical audit to illustrate compliance with audit standards,
however there are certain circumstances where one or the other is better.

• Pie charts - Show proportion, e.g. percentage compliance with standard. A pie chart would be
inappropriate if, for example, it was being used to illustrate ‘type of treatment’ in cases where patients
are likely to have had more than one treatment. In this instance the pieces of the pie would add up to
the total number of ‘treatments’ rather than the total number of patients, which could be misleading
and is not very meaningful in itself.
• Bar/column charts - Generally used to show frequency, e.g. number of patients seen by different staff:
nurse, Specialist Registrar, Consultant, etc. For example if the audit criteria stated that all patients seen
in A&E meeting certain criteria should be seen by a consultant, you might want to show what grade of
staff saw the patients if it was not the consultant.
• Versions of Bar charts (stacked or comparative) - Show more than one standard/ question per chart, e.g.
multiple results on one graph, bars divided to show percentages, etc.

EXAMPLE 3: Pie Charts, Bar Graphs & Comparative Bar Graphs

Was Standard 1 met? Was Standard 1 met? n = 50

Partially 30
20%
Number of patients

Fully
50% 25
25
20
15
15
10
10
5
Not met 0
n = 50 30% Fully Partially Not met

 2009 UHBristol Clinical Audit Team – Version 3 Page 5 of 8

 How To: Analyse & Present Data

Were the standards met? Fully Were the standards met? n = 50

40 n = 50 Partially 100%
90%

Percentage of patients
Number of patients

35 Not met Not met

80%
30 70%
25 60%
20 50% Partially
15 40%
30%
10
20% Fully
5 10%
0 0%
Standard 1 Standard 2 Standard 3 Standard 1 Standard 2 Standard 3

CREATING GOOD CHARTS

Good charts should focus on getting your message across, rather than creating fancy, and distracting,
images. Clutter should be avoided and the charts clearly labelled.

REMEMBER: Do not use graphs just for the sake of it.

EXAMPLE CHART 1: 37/40 members of staff (93%) took personal protective equipment on domiciliary
visits
Did member of staff take personal protective
equipment with them?

Yes

No
37

A chart might be considered unnecessary to illustrate the above data. It is always important to consider
whether or not people need to see a graphical representation of the data, in some cases simply expressing
the data as ‘x/n (y%)’ is sufficient. In a project with a lot of standards producing a chart for every single one
may confuse rather than clarify the results. People may not remember which image related to which
standard.

 2009 UHBristol Clinical Audit Team – Version 3 Page 6 of 8

 How To: Analyse & Present Data

EXAMPLE CHART 2: What’s wrong here?

What’s wrong?
• The 3D graph often leads to misleading
Standard 1
graphs: because the bars are not up against
back wall of chart, you could mis-read 25 and
15 as 24 and 14 or lower.
25
• The title needs more detail.
• The axes should be labelled. In this case the x 20

axis might not need labelling according to 15 Standard 1

what other titling is put on chart, but the y 10

axis needs to be labelled to be meaningful.
5
• In this instance the legend is not needed, as
0
there is only one data series. Yes No

• There is a lot of white space on the page. It

Standard 1: Head and Neck cancer patients should receive
looks unprofessional. a dental screen prior to radiotherapy
Revised version 30
25 n = 40
• It has been changed to 2D chart. 25

Number of patients
• Titles have been added. 20
• The format of gridlines and background has 15
15
been amended to add clarity.
10
• The sample size number has been added for
quick reference. 5

• There is better use of the space available. 0

Yes No
• The scale has been extended a little way past Was standard met?
highest bar.

SUMMARY

• Consider whether or not you need to represent your data graphically to get your message across.
• Consider which type of chart will deliver your message in the clearest way.
• Consider what information you should include in order to answer your audience’s questions. For
example, sample size, percentages if displaying real values (or vice versa). Importantly, avoid clutter.

CONTACT DETAILS/ USEFUL INFORMATION

CLINICAL AUDIT
• The UHBristol Clinical Audit website is available [online] via: http://www.uhbristol.nhs.uk/healthcare-
professionals/clinical-audit.html
• Contact details for the UHBristol Clinical Audit Team are available from the Clinical Audit Central Office
or [online] via: http://www.uhbristol.nhs.uk/healthcare-professionals/clinical-audit/contacts.html
• The full range of UHBristol ‘How To’ guides are available [online] via:
http://www.uhbristol.nhs.uk/healthcare-professionals/clinical-audit/how-to-guides.html
• A copy of the UHBristol Proposal Form, Presentation Template, Report Template, Summary Form, and
Action Form are available [online] via: http://www.uhbristol.nhs.uk/healthcare-professionals/clinical-
audit/doing-projects-at-ubht.html
• The UHBristol Clinical Audit Central Office can be contacted on tel. (0117) 342 3614 or e-mail:
[email protected]
• Clinical Audit Training Workshops can be booked through the Clinical Audit Central Office.

CLINICAL EFFECTIVENESS
• For advice on Clinical Effectiveness, including how to write guidelines, contact James Osborne, Clinical
Effectiveness Co-ordinator, tel. (0117) 342 3753 or e-mail: [email protected]

 2009 UHBristol Clinical Audit Team – Version 3 Page 7 of 8

 How To: Analyse & Present Data

PATIENT ENGAGEMENT
• For advice on Patient Involvement, including designing structured surveys and questionnaires contact
Paul Lewis, Patient Involvement Facilitator, tel. (0117) 342 3638 or e-mail: [email protected]
• For advice on Patient Involvement, including unstructured surveys and focus groups contact Tony
Watkin, Public Involvement Lead, tel. (0117 342 3729 or e-mail: [email protected]
• Surveys MUST be approved by the Trust's Questionnaire, Interview and Survey (QIS) Group. Proposals
should be submitted to Paul Lewis using the QIS proposal form. The proposal form is available [online]
via http://www.uhbristol.nhs.uk/healthcare-professionals/clinical-audit/doing-projects-at-ubht.html
• A copy of the UHBristol Covering Letter template is available [online] via the internal intranet site
http://connect/Governance/patientexperience/ppi/Pages/QISGroup.aspx

RESEARCH
• For advice on research projects contact the Research & Development Department, tel. (0117) 342 0233
or e-mail: r&[email protected]

LITERATURE REVIEWS
• For advice on literature reviews contact the Learning Resource Centre, tel. 0117 342 0105 or e-mail:
[email protected]

SAMPLE SIZES
• The Sample Size Calculator is available [online] via: http://www.uhbristol.nhs.uk/healthcare-
professionals/clinical-audit/how-to-guides.html

 2009 UHBristol Clinical Audit Team – Version 3 Page 8 of 8