Diabetes and Cardiovascular Events in Women With PCOS
Diabetes and Cardiovascular Events in Women With PCOS
A departmental clinical information system (Leicester Clinical Workstation) contains detailed information on diagnosis, investigations and treatment [encoded using the Clinical Terms v3 (Read Codes)] on all patients seen since 1988 in the Endocrinology service at the Leicester Royal Infirmary, United Kingdom, the main specialist endocrine provider for Leicestershire which covers a population of almost 1 million people. Data are collected and maintained as part of routine clinical practice, and since the early 1990s, the system has been used as the primary source of clinical information during every clinical contact with the patient. The 'problem list' data structure allows entry of any clinical concept which the clinician seeing the patient feels to be clinically relevant, but does not insist on completion of any specific restricted data set. We searched the database for all patients who had a recorded diagnosis of 'PCOS' at any time. For all the patients, the same report retrieved data on presence, dates and details of (i) symptoms and signs of PCOS, (ii) diabetes mellitus, (iii) other conditions that would exclude the patient from analysis (Cushing's syndrome and congenital adrenal hyperplasia), (iv) any type of CV disease or CV risk factor (including hypertension and smoking), (v) ethnic origin and (vi) death (for details of search strategy see Appendix 1).
The inclusion criteria for this cohort were therefore women who attended clinic between October 1988 and 1 November 2009 (point of data extraction) and were assigned a clinical diagnosis of PCOS and were at least 16 years of age at the time of data extraction. The majority (82·6%) of the patients had been reviewed under the care of one senior endocrinologist (TAH) over a period of over 20 years. Presenting symptoms such as hirsutism, infertility or menstrual irregularity were typically recorded, but the precise diagnostic criteria leading to the diagnosis were not systematically recorded prospectively. The consensus diagnostic criteria for PCOS also changed over the same time period. We therefore accepted the recorded diagnosis of PCOS in our database as evidence that an appropriate diagnosis of PCOS had been made in this patient based on prevailing clinical criteria at the time. Diagnostic criteria for PCOS were however reviewed in the available clinic letters and updated in the analysis database. As data were all recorded in the database as a part of routine clinical practice and as a clinical diagnosis of PCOS had been applied to the patient in the endocrinology clinic where patients were consistently reviewed by a senior endocrinologist, we considered that in patients where no symptoms were recorded, this was most likely evidence of failure to record relevant clinical findings rather than evidence of absence of these findings. The diagnostic criteria for PCOS were hyperandrogenism (presence of hirsutism, acne, androgenic alopecia or abnormal blood test indicating high total testosterone or raised free androgen index) and anovulation (oligomenorrahoea, amenorrhoea, infertility). During the period in which most of these patients presented, it was not the clinical policy of the department to perform routine ovarian ultrasound when the diagnosis of PCOS was clear on clinical and biochemical grounds. Based on this enhanced data set, subgroup analysis was performed for those with evidence for definite PCOS (two or more Rotterdam criteria (which given lack of ultrasound data in most cases represented the two main NIH criteria for PCOS).
Anonymized data on eligible patients with PCOS were extracted, and National Health Service (NHS) number and date of birth data were linked to the local hospital admission episodes database maintained by the Leicestershire NHS Health Informatics Services (HIS) database, to obtain information on CV events. The HIS database covers the same stable population of approximately 1 million individuals served by the endocrinology clinics and has been routinely linked to the local population register since 1985. All local hospital admissions for our population of women with PCOS were available since 4 April 1995. Relevant codes using International Classification of Disease: ICD-10 or ICD-9 and operating procedure codes were extracted for all patients, including codes for MI and angina (I20–I25 and 410–4140), heart failure (HF; I50 and 428–4289), Stroke (I60–I69 and 430–4371), angioplasty and angiography. For each patient, only the first event of each type of event was recorded in the analysis.
Date and diagnosis of diabetes was confirmed using clinical letters on the database or earliest abnormal blood test in hospital laboratory system. In view of a high number of people being misclassified and recent guidelines in the Classification of Diabetes, all of the patients with a recorded diagnosis of Type 1 diabetes were reviewed by an independent Diabetologist and their diagnosis on the database was changed to T2DM if they met the current classification criteria.
A categorical diagnosis of 'body weight' as normal, overweight or obese was used for all the patients with available related data by either categorizing a recorded BMI or using a recorded diagnosis of 'overweight' or 'obese' from the clinical database. BMI value was not always recorded in the database, particularly in the earlier years being studied, but it was longstanding departmental clinical practice to record a coded clinical diagnosis of weight classification using 'Overweight' (Read Code: X40YL) when BMI was >25–30 and 'Obesity' (C380.) when BMI was >30 and 'Morbid Obesity' (X40YQ) when BMI was >40; we therefore analysed these encoded categorical classifications of weight when a BMI value was not available.
Statistical analysis was performed using the spss 18 package (IBM Company, Armonk, NY, USA). The outcomes of interest were T2DM, MI, angina, HF, stroke and CV death. Additionally, in some analyses, we used a composite CV outcome, defined as any of MI, angina, HF, stroke and CV mortality.
The incidence rate of each of these outcomes was calculated as the number of new cases divided by the total number of person-years at risk for that condition. Individuals started contributing person-years from their first clinic visit or earliest registration on local HIS, whichever was last, if they were free of the disease of interest at that time. Registration on HIS was required because CV incident events were extracted from that data source. Individuals were then followed up until they had the event of interest, died, migrated out of the Leicestershire area or reached the end of observation (1 November 2009), whichever occurred first. If someone who had previously migrated was re-registered, then each period in the database was managed separately to the above end points.
Prevalence at the end of the observation period was estimated for each of the outcomes among all of the study participants and separately in four age groups (15–44, 45–54, 55–64 and ≥65 years) based on their age at 1 November 2009. Prevalence was defined as having a history of the event of interest by the end of the observation period. For example, prevalent MI means that the patient had had a MI at any point up to 1 November 2009.
We then investigated risk factors for T2DM and the composite CV outcomes using multiple logistic regression. The explanatory variables used in the model were age, body weight, Index of Multiple Deprivation as a marker of socio-economic status, hyperandrogenism, anovulation, ethnicity, smoking and history of hypertension (and diabetes status when analysing the composite CV outcome).
Odds ratios were used to compare the proportion of women with an event of interest at the end of observation with comparable data in the local and national female population. Local data were available from the HIS database calculated based on the number of the hospital admissions for that event of interest over the population denominators for that age group. National data were available from the Health Survey for England.
Sensitivity analyses were performed by repeating all analyses without those women who were not classified as definitely having PCOS according to Rotterdam criteria.
For all the categorical data such as signs, symptoms, smoking and history of hypertension, only positive response were recorded in the database and therefore missing fields could relate to either a negative response or missing data. We did not replace missing data for any of the variables, except ethnicity, which was not recorded for a minority of the population (10%) in whom an expert administrator reviewed the names and assigned the ethnicity on that basis.
All P-values shown are two-sided, and statistical significance was assessed at the 5% level.
Methods
Study Population
A departmental clinical information system (Leicester Clinical Workstation) contains detailed information on diagnosis, investigations and treatment [encoded using the Clinical Terms v3 (Read Codes)] on all patients seen since 1988 in the Endocrinology service at the Leicester Royal Infirmary, United Kingdom, the main specialist endocrine provider for Leicestershire which covers a population of almost 1 million people. Data are collected and maintained as part of routine clinical practice, and since the early 1990s, the system has been used as the primary source of clinical information during every clinical contact with the patient. The 'problem list' data structure allows entry of any clinical concept which the clinician seeing the patient feels to be clinically relevant, but does not insist on completion of any specific restricted data set. We searched the database for all patients who had a recorded diagnosis of 'PCOS' at any time. For all the patients, the same report retrieved data on presence, dates and details of (i) symptoms and signs of PCOS, (ii) diabetes mellitus, (iii) other conditions that would exclude the patient from analysis (Cushing's syndrome and congenital adrenal hyperplasia), (iv) any type of CV disease or CV risk factor (including hypertension and smoking), (v) ethnic origin and (vi) death (for details of search strategy see Appendix 1).
The inclusion criteria for this cohort were therefore women who attended clinic between October 1988 and 1 November 2009 (point of data extraction) and were assigned a clinical diagnosis of PCOS and were at least 16 years of age at the time of data extraction. The majority (82·6%) of the patients had been reviewed under the care of one senior endocrinologist (TAH) over a period of over 20 years. Presenting symptoms such as hirsutism, infertility or menstrual irregularity were typically recorded, but the precise diagnostic criteria leading to the diagnosis were not systematically recorded prospectively. The consensus diagnostic criteria for PCOS also changed over the same time period. We therefore accepted the recorded diagnosis of PCOS in our database as evidence that an appropriate diagnosis of PCOS had been made in this patient based on prevailing clinical criteria at the time. Diagnostic criteria for PCOS were however reviewed in the available clinic letters and updated in the analysis database. As data were all recorded in the database as a part of routine clinical practice and as a clinical diagnosis of PCOS had been applied to the patient in the endocrinology clinic where patients were consistently reviewed by a senior endocrinologist, we considered that in patients where no symptoms were recorded, this was most likely evidence of failure to record relevant clinical findings rather than evidence of absence of these findings. The diagnostic criteria for PCOS were hyperandrogenism (presence of hirsutism, acne, androgenic alopecia or abnormal blood test indicating high total testosterone or raised free androgen index) and anovulation (oligomenorrahoea, amenorrhoea, infertility). During the period in which most of these patients presented, it was not the clinical policy of the department to perform routine ovarian ultrasound when the diagnosis of PCOS was clear on clinical and biochemical grounds. Based on this enhanced data set, subgroup analysis was performed for those with evidence for definite PCOS (two or more Rotterdam criteria (which given lack of ultrasound data in most cases represented the two main NIH criteria for PCOS).
Anonymized data on eligible patients with PCOS were extracted, and National Health Service (NHS) number and date of birth data were linked to the local hospital admission episodes database maintained by the Leicestershire NHS Health Informatics Services (HIS) database, to obtain information on CV events. The HIS database covers the same stable population of approximately 1 million individuals served by the endocrinology clinics and has been routinely linked to the local population register since 1985. All local hospital admissions for our population of women with PCOS were available since 4 April 1995. Relevant codes using International Classification of Disease: ICD-10 or ICD-9 and operating procedure codes were extracted for all patients, including codes for MI and angina (I20–I25 and 410–4140), heart failure (HF; I50 and 428–4289), Stroke (I60–I69 and 430–4371), angioplasty and angiography. For each patient, only the first event of each type of event was recorded in the analysis.
Date and diagnosis of diabetes was confirmed using clinical letters on the database or earliest abnormal blood test in hospital laboratory system. In view of a high number of people being misclassified and recent guidelines in the Classification of Diabetes, all of the patients with a recorded diagnosis of Type 1 diabetes were reviewed by an independent Diabetologist and their diagnosis on the database was changed to T2DM if they met the current classification criteria.
A categorical diagnosis of 'body weight' as normal, overweight or obese was used for all the patients with available related data by either categorizing a recorded BMI or using a recorded diagnosis of 'overweight' or 'obese' from the clinical database. BMI value was not always recorded in the database, particularly in the earlier years being studied, but it was longstanding departmental clinical practice to record a coded clinical diagnosis of weight classification using 'Overweight' (Read Code: X40YL) when BMI was >25–30 and 'Obesity' (C380.) when BMI was >30 and 'Morbid Obesity' (X40YQ) when BMI was >40; we therefore analysed these encoded categorical classifications of weight when a BMI value was not available.
Statistical Analysis
Statistical analysis was performed using the spss 18 package (IBM Company, Armonk, NY, USA). The outcomes of interest were T2DM, MI, angina, HF, stroke and CV death. Additionally, in some analyses, we used a composite CV outcome, defined as any of MI, angina, HF, stroke and CV mortality.
The incidence rate of each of these outcomes was calculated as the number of new cases divided by the total number of person-years at risk for that condition. Individuals started contributing person-years from their first clinic visit or earliest registration on local HIS, whichever was last, if they were free of the disease of interest at that time. Registration on HIS was required because CV incident events were extracted from that data source. Individuals were then followed up until they had the event of interest, died, migrated out of the Leicestershire area or reached the end of observation (1 November 2009), whichever occurred first. If someone who had previously migrated was re-registered, then each period in the database was managed separately to the above end points.
Prevalence at the end of the observation period was estimated for each of the outcomes among all of the study participants and separately in four age groups (15–44, 45–54, 55–64 and ≥65 years) based on their age at 1 November 2009. Prevalence was defined as having a history of the event of interest by the end of the observation period. For example, prevalent MI means that the patient had had a MI at any point up to 1 November 2009.
We then investigated risk factors for T2DM and the composite CV outcomes using multiple logistic regression. The explanatory variables used in the model were age, body weight, Index of Multiple Deprivation as a marker of socio-economic status, hyperandrogenism, anovulation, ethnicity, smoking and history of hypertension (and diabetes status when analysing the composite CV outcome).
Odds ratios were used to compare the proportion of women with an event of interest at the end of observation with comparable data in the local and national female population. Local data were available from the HIS database calculated based on the number of the hospital admissions for that event of interest over the population denominators for that age group. National data were available from the Health Survey for England.
Sensitivity analyses were performed by repeating all analyses without those women who were not classified as definitely having PCOS according to Rotterdam criteria.
For all the categorical data such as signs, symptoms, smoking and history of hypertension, only positive response were recorded in the database and therefore missing fields could relate to either a negative response or missing data. We did not replace missing data for any of the variables, except ethnicity, which was not recorded for a minority of the population (10%) in whom an expert administrator reviewed the names and assigned the ethnicity on that basis.
All P-values shown are two-sided, and statistical significance was assessed at the 5% level.
Source...