Introduction Approximately 366 million people worldwide live with diabetes and this figure is expected to rise. Among the correct diagnosis, there will be errors in the diagnosis, classification and coding, resulting in adverse health and financial implications.
Aim To determine the prevalence and characteristics of diagnostic errors in people with diabetes managed in primary care settings.
Methods We conducted a cross-sectional study in nine general practices in Leicester, UK, from May to August 2011, using a validated electronic toolkit. Searches identified cases with potential errors which were manually checked for accuracy.
Results There were 54 088 patients and 2434 (4.5%) diagnosed with diabetes. Out of 316 people identified with potential errors with the toolkit, 180 (57%) had confirmed errors after manually reviewing the records, resulting in an error prevalence of 7.4%. Correctly coded people on registers had significantly greater glycated haemoglobin (HbA1c) reductions. There were no significant differences between patients with and without errors in their HbA1C, body mass index, age and size of practice. There was also no significant association of the errors with pay-for-performance initiatives; however, those patients not on disease register had worse glycaemic control.
Conclusions A high prevalence of diabetic diagnostic errors was confirmed using medication, biochemical and demographic data. Larger studies are needed to more accurately assess the scale of this problem. Automation of these processes might be possible, which would allow searches to be even more user friendly.
- DIABETES & ENDOCRINOLOGY
Statistics from Altmetric.com
Globally, it is estimated that 366 million people are living with diabetes.1 As the proportion of diagnosed cases of diabetes increases, it is plausible to suggest that there may also be a similar increase in the cases of diabetes errors. The three types of errors described are misclassification, miscoding and misdiagnosis. Miscoding occurs when a vague diagnostic code is used, making it impossible to determine the type of diabetes. Misclassification refers to giving the patient a type of diabetes classification he or she did not have. Misdiagnosis refers to the situation where the patient is not diabetic but has been labelled as such.2 The logic behind these definitions is based on the reasoning that, first, people with type 2 diabetes mellitus (T2DM) are misclassified as having type 1 diabetes mellitus (T1DM) if they are not on insulin. People with T1DM require insulin for survival and so everyone with T1DM should be prescribed insulin from diagnosis.3 Second, if people are labelled as T1DM and are prescribed both insulin and oral hypoglycaemic agents (OHAs), they are more likely to have T2DM except for the small minority of patients with T1DM and obesity.4 Last, T2DM patients are also likely to be misclassified as having T1DM if they are started on OHAs before insulin. Sometimes, it can be extremely difficult to differentiate between T1DM and T2DM as the former may start in adult life in someone with a T2DM phenotype.
Similarly, T1DM is probably incorrectly classified as T2DM if insulin was required from or within 6 months of diagnosis. People with T1DM will have continuous therapy with insulin from diagnosis. Blood glucose tests and glycated haemoglobin (HbA1c) levels are needed for diagnosis of diabetes. If at diagnosis, the blood glucose levels are less than 7 mmol/L or the HbA1c levels are less than 6.5% or <48 mmol/mol and the patients were labelled as T2DM but not on any treatment, they are probably misdiagnosed.5
Due to lack of homogeneity, a recent systematic review could not ascertain the prevalence of these errors. This review only identified studies dealing with classification, diagnostic errors and problems around maturity-onset diabetes of the young (MODY), latent autoimmune diabetes in adults and pancreatic diabetes.2 The implications of wrong diagnosis, coding or classification can affect optimal treatment regimen and cause inappropriate financial and psychological impact in such patients. In addition, these data may not be reliable for research or audit purposes.
A recent evaluation found that people with T2DM misclassified as T1DM tended to be older (mean 62 vs 47 years old). People misdiagnosed as having T2DM have apparently ‘excellent’ glycaemic control with mean HbA1c 5.3% (34 mmol/mol) versus 7.2% (55 mmol/mol) (p<0.001). People with vague codes not included in the pay-for-performance (P4P) register (miscoded) have worse glycaemic control (HbA1c 8.1% (65 mmol/mol) vs 7.0% (53 mmol/mol), p=0.006).6 This evaluation followed a small pilot study looking at five practices in the southeast of England which identified a 5% prevalence of errors in the coding, classification and diagnosis of diabetes.7 To date, no such studies have taken place outside London and the southeast of England where the demography and organisation of diabetes care delivery may be different.
This study therefore sets out to determine the prevalence and characteristics of diagnostic errors in people with diabetes managed in primary care settings by the use of an electronic toolkit on practice databases. The study was carried out in an inner city location in mid-England. The toolkit was developed based on the logical reasoning underlying the definitions of the various errors as described above. Since the outlined reasons underlying the definitions of the various errors are only a guide, manual assessment of the outputs produced by the electronic toolkit is essential to further refine the data. This is achieved by ensuring that no particular patient is analysed twice or more and by correcting the suspected errors in the diagnosis, coding or classification suggested by the toolkit.
We carried out a cross-sectional study in inner city Leicester, UK, using anonymised data comprised of patient demographic and laboratory test data like blood glucose and HbA1c results which were available on practice computer databases. Other available information included diagnostic codes, dates of first diagnosis, therapeutic data including insulin, metformin and other OHAs. These live data were extracted between May and August 2011 using the toolkit which was easily downloaded (http://www.clininf.eu/diabetes.html). The toolkit was developed using logical models to phrase queries using the diagnostic rules within the algorithm for the diagnosis of diabetes. The detailed description of its developmental process has been outlined elsewhere.7
The review was conducted in accordance with best practice by a team responsible for the care of these patients. For each patient identified, an output sheet was prepared for the responsible clinician. Ethics approval was sought from the local research and governance team but they felt it was not necessary to get ethical approval as this was a service evaluation in which there was no direct patient contact from third parties and the analysis was all done on unidentifiable patient data output. For the majority of patients, the errors identified were simple to correct and the responsible clinicians made the necessary changes to the patient's computer record. For a small minority of patients for whom clinicians were unable to correct the errors, they had the chance to speak to the lead diabetes clinician.
In 2009/10, there were 18 998 people aged 17 years and older diagnosed with diabetes in Leicester city. For adult patients’ aged 17 years and older in Leicester city, the prevalence of confirmed diabetes was 6.9% at the time of the search. The age structure of Leicester is skewed towards a relatively young population as 45% of residents are below 29 years of age. The over 60 population in Leicester is declining as older residents move to the neighbouring areas in the countryside. Women make up 52% of Leicester population as against 48% for men.8
The study was carried out in nine general practices from the 15 practices in a cluster of practices in one locality. Since 2012, diabetes care in Leicester has been undergoing transformation and the proposed model which is currently being piloted separates the complexity of diabetes care to a more complex care which requires specialist input and less complex care provided in primary care closer to the patients’ home, that is, at a general practitioner practice. Thus, primary care provides core and enhanced services with specialist support depending on what the practices provide and secondary care provides specialist care in community or hospital settings (figure 1).9
The numbers and percentages of patients misclassified (either as T1DM or T2DM), misdiagnosed or miscoded were evaluated to establish prevalence of the errors. Patients’ ages, earliest and most recent recorded body mass index (BMI) and HbA1cs were also retrieved to help characterise the coding, diagnostic and classification errors. The results of the analyses are presented using the descriptive statistics of the variables and the detailed results of the various analytic statistical models used. Statistical analyses were performed using SPSS (V.18.0, Chicago, Illinois, US). Continuous (HbA1c, age and BMI) data were expressed as mean and SD if normally distributed. Categorical data were expressed as number followed by percentage. To compare variables between the two groups, independent samples t tests were used for normally distributed continuous variables, and χ2 tests for categorical variables. A p value<0.05 is considered statistically significant (two-sided). For the analytical modelling, the categorical variables looked at were preaudit diagnosis of diabetes, postaudit diagnosis of diabetes, type of diabetes, whether or not there was a change in diagnosis of diabetes and size of practice. As P4P process was introduced in the UK in 2004, we examined records prior to and after 2004.
The practices had a combined list size of approximately 54 088, out of which 2434 had diabetes (4.5%). The searches identified 316 people with possible errors of misdiagnosis, misclassification and miscoding from the total of 2434 patients with diabetes. After manually reviewing the data for the 316 people identified with potential errors, 180 (57%) had confirmed errors, meaning that the proportion of all the 2434 diabetes patients with some form of errors in coding, classification or diagnosis in this cohort was 7.4%. Of the 180 patients identified, 49 (27.2%) were misclassified as T1DM, 34 (18.9%) were misclassified as T2DM and 86 (47.8%) were miscoded (table 1). The number that did not have diabetes at all but were labelled as such was 11 (6.1%). Sixteen patients with T1DM actually had T2DM and 13 patients with T2DM had T1DM.
Characteristics of the raw toolkit output before the manual review
Of the 316 cases identified by the toolkit with possible errors, 179 (56.6%) were men. The mean age was 57.1 years (SD 19.0). People in the middle aged groups and the elderly population were more represented in this cohort than people less than 35 years of age. The mean earliest recorded BMI was 28.5 kg/m2 (SD 6.6). The mean earliest recorded HbA1c was 8.6% (SD 2.7) (70 mmol/mol). Patients coded as having T2DM constituted the majority of the cases identified before the review, accounting for 124 (39.2%) of cases. T1DM accounted for 93 (29.6%) of cases, while those with vague diagnostic codes accounted for 89 (28.3%). Only 3 (1.0%) cases were labelled as secondary diabetes and 2 (0.6%) had no label at all. Gestational diabetes accounted for 5 (1.6%) of the cases identified (table 2).
Characteristics of the output after the manual review
After the review, 160 cases were recoded or reclassified correctly based on the predefined query definitions. For the remaining 20 cases, decision about the correct diagnosis or coding was made using available data on age at diagnosis, family history and BMI and not necessarily the stipulated queries. Overall, 72 (45.0%) cases were reclassified as T2DM and 32 (20.0%) as T1DM. In all, 44 (27.50%) remained unclassified and needed further investigation to correctly classify them (table 3).
Those patients who had errors had a higher level of mean HbA1c of 8.5% (SD 2.34) (69 mmol/mol) than those who had no errors (mean HbA1c of 8.2% (SD 2.03) (66 mmol/mol)) but this was not statistically significant (mean difference=−0.26 95% CI −0.78 to 0.25; p=0.32) (table 4). People with correct codes and diagnosis on disease registers had significantly greater HbA1c reductions (mean earliest recorded HbA1c of 8.7% decreasing to a current mean recorded HbA1c of 8.1% paired t test p=0.032).
True T2DM people tended to have significant decreases in HbA1C levels, mean earliest recorded HbA1c of 8.7% versus mean current recorded HbA1c of 8.1% (paired t test p=0.032). Their mean BMI also significantly increased from 28.5 kg/m² to 29.3 kg/m² (paired t test p=0.02). T2DM patients misclassified as T1DM tended to be older than the true T1DM patients but this was not statistically significant; mean age 46.4 versus 44.4 (independent t test p=0.56). They have lower HbA1c than true T1DM, but this was not statistically significant; 8.4% versus 9.2% (independent t test p=0.11).
There was no difference in terms of the mean age of those patients who had errors and those who did not, size of practice and those diagnosed after the introduction of the P4P (p=0.055, p=0.58, p=0.24, respectively). People miscoded and so not on disease registers were found to have higher HbA1c levels than those without errors and on disease registers but this was not statistically significant (HbA1c 8.5% vs 8.2%, p=0.372).
This study demonstrates that the prevalence of errors in people with diabetes in primary care was 7.4% following the review, slightly more than the 5% recorded in the pilot,7 thus suggesting that the problem of diagnostic and classification errors could be bigger. Overall, 45.0% were reclassified as T2DM, 20.0% as T1DM and 27.5% remained unclassified. Our study suggests that the audit toolkit, when used in primary care to perform such searches, can detect the burden of reporting errors. People with correct codes and diagnosis on disease registers have significant improvements in their HbA1c. Previous studies have looked at the prevalence of errors in classification, coding and diagnosis of diabetes;6 ,7 this study further confirms and highlights the characteristics of this group of patients. There was a trend for people with errors to have poorer glycaemic controls. Correctly identified T2DM patients tended to have significant decreases in HbA1C levels but their BMIs increased with time.
T1DM diabetes accounts for only about 5% of all diabetes patients,10 yet there were more errors relating to it in this study, suggesting that characterising patients as T1DM at diagnosis is much more challenging than in the case of T2DM. Just over a quarter of the patients needed further investigations to categorise their diagnosis and may need specialist input and tests but these may not be available readily. It is estimated that more than 80% of MODY cases in the UK are currently misdiagnosed as T1DM or T2DM.11 It is highly unlikely that the diagnosis of MODY can even be suspected in primary care where most people with diabetes are diagnosed and managed.12 Recently, the UK diagnostic testing centre reported a delay of about 13 years in the diagnosis of MODY and reported a wide variation of referral rates across the country.11 In addition, T2DM is commonly being diagnosed at younger ages due to the obesity epidemic13 causing more diagnostic uncertainties.
Errors in diagnosis or coding or classification appear to lead to suboptimal management and hence worse HbA1c levels. People with T1DM will normally be started on insulin, even if the diagnosis of T1DM was an error.
The introduction P4P in the UK in 2004 has been shown in various observational studies to drive improvement in the care of diabetes.14 The data in this study were analysed to see if there was any difference in the proportion of patients with errors in coding, classification and diagnosis of diabetes in those diagnosed before 2004 and after 2004. There was no statistical difference in the two groups. So there may not be a link between the aggressive management of diabetes to achieve P4P targets and errors in coding, classification or diagnosis. This is the first study to suggest this, but those patients not on disease registers tended to have a worse glycaemic control.
Where patients present for the first time with diabetes complications and require emergency treatment, and diagnosis of type of diabetes is not obvious, clinicians should avoid labelling these patients with a type of diabetes unless diagnosis is clearly obvious. Further investigations in diabetes specialist clinics are advisable, where detailed history of diabetes in the family, antibody testing, genetic test and c-peptide testing can be done.
Even though the toolkit is very useful, there are still some inherent weaknesses in it since only 180 of the total 316 patients with potential errors had confirmed errors after the review. This study was only conducted in nine practices in one large inner city, and the results can therefore not be generalised. In addition, there is a possibility of a selection bias as the nine practices were the ones that agreed to take part out of the 16 in the cluster. The large numbers in the population however minimise this potential limitation.
The searches found a higher rate of underlying errors (57%) compared with those found in the previous pilot, in which 42% had coding errors.15 Larger and more methodologically robust studies are needed to more accurately assess the scale of this problem. With further research, automation of these processes might be possible, which would allow searches to be even more user friendly. Algorithms that can be processed by machines appear to achieve similar results.16 There is also the need to consider similar toolkits for patients at risk of developing diabetes.
Errors in coding, classification or diagnosis in a cohort of people with diabetes will affect the care of these individuals. Patients with correct codes and diagnosis on disease registers achieve significant improvements in their glycaemic control.
The main message from this study is that the prevalence of the errors in coding classification and diagnosis of diabetes is 7.4%.
There was a trend for people with errors to have poorer glycaemic controls.
People with correct codes and diagnosis on disease registers had significantly greater glycated haemoglobin (HbA1c) reductions.
Correctly identified type 2 diabetes mellitus patients tended to have significant decreases in HbA1C levels but their body mass indexes increased with time.
There may not be a link between the aggressive management of diabetes to achieve pay-for-performance targets and errors in coding, classification or diagnosis.
Current research questions
Larger and more methodologically robust studies are needed to more accurately assess the scale of this problem.
With further research, automation of these processes might be possible, which would allow searches to be even more user friendly.
There is also a need to consider similar toolkits for patients at risk of developing diabetes.
Stone MA, Camosso-Stefinovic J, Wilkinson J, et al. Incorrect and incomplete coding and classification of diabetes: a systematic review. Diabet Med 2010;27:491–7.
Sadek NH, Sadek AR, Tahir A, et al. Evaluating tools to support a new practical classification of diabetes: excellent control may represent misdiagnosis and omission from disease registers is associated with worse control. Int J Clin Pract 2012;6:874–82.
de Lusignan S, Khunti K, Belsey J, et al. A method of identifying and correcting miscoding, misclassification and misdiagnosis in diabetes: a pilot and validation study of routinely collected routinely collected data. Diabet Med 2010;27:203–9.
We acknowledge the funding received to support this project from the Leicester Primary Care Group Cluster, and The National Institute for Health Research Collaboration for Leadership in Applied Health Research and Care—Leicestershire, Northamptonshire and Rutland (NIHR CLAHRC—LNR).
Contributors SS led the design, analysis and interpretation of data and drafted the article. SM revised the article for important intellectual content. MJD, SdeL and KK were involved in the conception and revised the article for important intellectual content. All authors gave final approval of the version to be published.
Funding Leicester City CCG. Leicester Primary Care Group.
Competing interests SS has received honoraria for serving on Advisory Boards for Novartis, and Novo. SM has received a training fellowship from Novo Nordisk UK research foundation. MJD has received funds for research, honoraria for speaking at meetings and has served on Advisory Boards for Lilly, Sanofi Aventis, MSD, Novo Nordisk, BMS, BI and Roche. KK has received funds for research, honoraria for speaking at meetings and/or served on Advisory Boards for Astra Zeneca, Lilly, Novartis, Pfizer, Servier, Sanofi Aventis, MSD and Novo Nordisk. SdL has no conflicts of interests. He was the informatics lead in a RCGP/NHS Diabetes Classification of Diabetes Task Group of which KK was chair.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.