Downloaded from http://bmjopen.bmj.com/ on June 18, 2017 - Published by group.bmj.com
PEER REVIEW HISTORY
BMJ Open publishes all reviews undertaken for accepted manuscripts. Reviewers are asked to
complete a checklist review form (http://bmjopen.bmj.com/site/about/resources/checklist.pdf) and
are provided with free text boxes to elaborate on their assessment. These free text comments are
reproduced below. Some articles will have been accepted based in part or entirely on reviews
undertaken for other BMJ Group journals. These will be reproduced where possible.
ARTICLE DETAILS
TITLE (PROVISIONAL)
AUTHORS
Undiagnosed diabetes from cross sectional GP practice data: An
approach to identify communities with high likelihood of undiagnosed
diabetes
Bagheri, Nasser; McRae, Ian; Konings, Paul; Butler, Danielle;
Douglas, Kirsty; Del Fante, Peter; Adams, Robert
VERSION 1 - REVIEW
REVIEWER
REVIEW RETURNED
GENERAL COMMENTS
Eva Vivian
University of Wisconsin-Madison
School of Pharmacy
USA
23-Apr-2014
This is a very important study that emphasized the need for
increased screening among the middle class or less disadvantaged
groups.
The authors acknowledge that only 35% of BMI values were
available. Therefore BMI values were imputed. It is important for the
authors to address the limitations of imputing these values (Since
the imputed observations are
themselves estimates, their values have
corresponding random error).
Strong work. This is a very important study that emphasized the
need for increased screening among the middle class or less
disadvantaged groups.
The authors acknowledge that only 35% of BMI values were
available. Therefore BMI values were imputed. It is important for the
authors to address the limitations of imputing these values (Since
the imputed observations are
themselves estimates, their values have
corresponding random error).
REVIEWER
REVIEW RETURNED
Michael Soljak
Department of Public Health & Primary Care
School of Public Health
Imperial College London
19-Jun-2014
GENERAL COMMENTS
The objectives of this study were to derive model-based estimates of
Downloaded from http://bmjopen.bmj.com/ on June 18, 2017 - Published by group.bmj.com
diabetes prevalence and to apply them to general practice data, to
explore the prevalence of undiuagnosed diabetes.
Introduction
The para beginning p4 line 46 fits better in the Methods.
Methods
BMI was only available for 36% of the
population, so was imputed. A general rule is that imputation should
not be used with more >50% missing data.
The method by which a diabetes diagnosis was made in NWAHS
data is not defined.
The steps involved in regression modelling and derivation of the
estimates are not explained. How were final variables selected in the
model? There is no explanation how model performance was
validated internally e.g. area under ROC curve/c statistics.
It is advantageous to use all available risk factors in the derivation
dataset as the performance of this model can be compared to one
which uses only the variables available in the local data used to
estimate prevalence.
Results
The results of the regression modelling should be presented in a
table with regression coefficents/odds ratios and ROC curves for
internal validation.
The sentence beginning p5 line 48 needs to be rewritten. In Table 1,
confidence intervals (CIs) are given for undiagnosed prevalence. As
these are modelled estimates derived from regression coefficients,
not a sample, it is not clear how they were derived.
The authors could test for spatial non-stationarity using a Moran's I
test, which is very simple to apply.
This paper potentially adds significantly to the current literature on
diabetes prevalence in small populations, which has not been
extensively investigated. However the methods, especially
regression modelling, need further documentation and this should be
reflected in additional results.
VERSION 1 – AUTHOR RESPONSE
Reviewer #1: Strong work. This is a very important study that emphasized the need for increased
screening among the middle class or less disadvantaged groups.
answer: Thank you
Reviewer #1 Reviewer #2: The authors acknowledge that only 35% of BMI values were available.
Therefore BMI values were imputed. It is important for the authors to address the limitations of
imputing these values. A general rule is that imputation should not be used with more than 50%
missing data.
We included the following sentences in last paragraph in the discussion area.
"While noting that there is a general rule of thumb that imputation should not be used for more than
half a population, as regression imputation was used we were in effect using a synthetic estimate of
Downloaded from http://bmjopen.bmj.com/ on June 18, 2017 - Published by group.bmj.com
the BMI where the original was not available, so while subject to variation the estimates were not
subject to the biases potentially inherent in PMM or hot-decking methodologies.”
Reviewer #2: This paper potentially adds significantly to the current literature on diabetes prevalence
in small populations, which has not been extensively investigated.
answer: Thank you
Reviewer #2: The para beginning on p4, line 46 fits better in methods
Answer:This paragraph was moved from introduction section to the beginning of the methods section.
(the first highlighted paragraph in methods section)
Reviewer #2: The method by which a diabetes diagnosis was made in NWAHS data is not defined
Answer: We included the following sentence in the main text in page 5, data source section:
“In the NWAHS, people with diagnosed diabetes were defined as those reporting that they had been
told by a doctor that they had diabetes.”
Reviewer #2: The steps involved in regression modelling and derivation of the estimates are not
explained. How were final variables selected in the model? There is no explanation how model
performance was validated internally e.g. area under ROC curve/c statistics.
Answer: In the model construction section, we explained the steps involved in logistic regression
modelling and that final variables were selected in the model based on availability and previous
studies. We performed a ROC analysis for our regression model and results are presented with Table
1 which we have added to the supplementary section.
Reviewer #2: It is advantageous to use all available risk factors in the derivation dataset as the
performance of this model can be compared to one which uses only the variables available in the
local data used to estimate prevalence.
Answer: In addition to the model reported in Table 1, we estimated a logistic regression model
including other available risk factors: waist circumference, marital status, work status, income,
education, cholesterol from the derivation dataset (NWAHS dataset). The area under the ROC was 81
using all available variables compared to 76 when using only the clinically available variables and the
difference was significant although small. However, among those new variables included in the new
regression model only waist and cholesterol were significant. We could not include these two
variables in the regression model used because of unavailability or high rate of missing values for
these variables in the clinical data.
Reviewer #2: The results of the regression modelling should be presented in a table with regression
coefficients/odds ratios and ROC curves for internal validation.
answer: The results of the logistic regression model with corresponding coefficients/odds ratios and
ROC/c statistics are now presented in Table 1 in the supplementary section. We included the
following results from regression model in the main text in the results section.
“The logistic regression model showed that those people who are male, over 50 years old, with higher
BMI and systolic blood pressure, more disadvantaged, pensioner and smoker have higher odds of
having diabetes (see Table 1, Supplementary).”
Reviewer #2: The sentence beginning p5 line 48 needs to be rewritten.
Answer: This sentence was rewritten in main text, in page 5, last paragraph as;
“The clinical data were drawn from a multisite GP practice in an area of Western Adelaide which
includes the LeFevre Peninsula.”
Reviewer #2: In Table 1, confidence intervals (CIs) are given for undiagnosed prevalence. As these
are modelled estimates derived from regression coefficients, not a sample, it is not clear how they
Downloaded from http://bmjopen.bmj.com/ on June 18, 2017 - Published by group.bmj.com
were derived.
Answer: We included the following sentence in result section before Table 2 in the main text.
“Confidence intervals for the predicted prevalences reflect the predictive accuracy of the logistic
regression model.”
Reviewer #2: The authors could test for the spatial non-stationary using Moran’s I test.
Answer: A very helpful suggestion. The aim for the next step of this project is to apply a
Geographically Weighted Regression (GWR) technique and Moran’s I method to investigate the
spatially varying relationships in diagnosed and undiagnosed diabetes in the study area. The result
will be demonstrated in another paper which authors are currently developing.
Downloaded from http://bmjopen.bmj.com/ on June 18, 2017 - Published by group.bmj.com
Undiagnosed diabetes from cross-sectional
GP practice data: an approach to identify
communities with high likelihood of
undiagnosed diabetes
Nasser Bagheri, Ian McRae, Paul Konings, Danielle Butler, Kirsty
Douglas, Peter Del Fante and Robert Adams
BMJ Open 2014 4:
doi: 10.1136/bmjopen-2014-005305
Updated information and services can be found at:
http://bmjopen.bmj.com/content/4/7/e005305
These include:
Supplementary Supplementary material can be found at:
Material http://bmjopen.bmj.com/content/suppl/2014/07/23/bmjopen-2014-005
305.DC1
References
This article cites 18 articles, 7 of which you can access for free at:
http://bmjopen.bmj.com/content/4/7/e005305#BIBL
Open Access
This is an Open Access article distributed in accordance with the Creative
Commons Attribution Non Commercial (CC BY-NC 4.0) license, which
permits others to distribute, remix, adapt, build upon this work
non-commercially, and license their derivative works on different terms,
provided the original work is properly cited and the use is
non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Email alerting
service
Receive free email alerts when new articles cite this article. Sign up in the
box at the top right corner of the online article.
Topic
Collections
Articles on similar topics can be found in the following collections
General practice / Family practice (629)
Health services research (1392)
Public health (2132)
Research methods (588)
Notes
To request permissions go to:
http://group.bmj.com/group/rights-licensing/permissions
To order reprints go to:
http://journals.bmj.com/cgi/reprintform
To subscribe to BMJ go to:
http://group.bmj.com/subscribe/