Estimating the workload associated with symptoms-based ovarian cancer screening in primary care: an audit of electronic medical records

We identified potential GP surgeries in London, Newbury, Wokingham and Bracknell via research contacts and collaborators. Of those willing to participate we only included those with a list size of at least 5500, that had been using EMIS (Egton Medical Information Systems one of the main software suppliers to UK general practices) LV (Limited Version) for at least one year, had at least one female GP (in case women are less likely to report gynaecological symptoms to a male GP), and no unusual demographic composition. The threshold for list size was based on the average of 5891 patients for England set by the 2004 General Medical Services contract [16].

Only anonymised data were used in the study, therefore ethics approval was not necessary. We used EMIS to download data on permanently registered women aged 45–74 years who presented over one week. Women whose records showed that they had a history of ovarian cancer or were actively receiving treatment for a malignancy (except for hormonal treatment only) were excluded.

Downloads took place between November 2007 and May 2008. We developed a search strategy in EMIS LV which downloaded data on women aged 45–74 who consulted over one week. Data included the woman’s age, medical history and details of the most recent previous consultation entries up to a maximum of 10. The week examined preceded the date of download by one day to allow time for after-hours and home visits to be entered.

We also collected the following details from each practice:

Consultation data in EMIS include entries that are not actual consultations such as incoming letters, administrative notes and laboratory test results. For each consultation entry we determined who was seen/encountered (from the name of the person who made the entry; eg a doctor or receptionist’s name), the consultation location (from the consultation type field; eg “GP surgery”, “path lab”). Consultation entries were then limited to those that were with a GP or a nurse and in-person (GP surgery or home visits) or over the telephone.

The same researcher (AWL) extracted symptoms from the consultation free text and clinical codes. Extracted symptoms were then coded as relevant or not relevant to ovarian cancer according to a pre-defined list of symptoms most commonly reported by women with ovarian cancer [2]-[10]. Symptoms included pelvic or abdominal pain or discomfort; abdominal bloating; increased abdominal size/abdominal distension; abdominal lump; indigestion; constipation; diarrhoea; nausea or vomiting; other gastrointestinal symptoms (eg wind, change in bowel habit); loss of appetite; weight loss, fatigue; urinary frequency or urgency; postmenopausal bleeding; and vaginal discharge. Vaginal discharge is not usually considered to be a typical ovarian cancer symptom, but was included because we found it to be independently associated with ovarian cancer in a case–control study [12]. A systematic coding frame (compiled with the input of two gynaecologists: Professor Usha Menon and Dr Aarti Sharma) was used to ensure symptoms were classified consistently.

Details of co-morbidities (type and date diagnosed/recorded) were obtained from the women’s medical history summary data.

We estimated the proportion of women who would be offered targeted ‘screening’ in GP primary care over one week under two different scenarios. The first used NICE guidelines which recommend CA125 testing in women who have any one of persistent abdominal distension, feeling full (early satiety) and/or loss of appetite, pelvic or abdominal pain or increased urinary urgency and/or frequency, on a persistent or frequent basis particularly more than 12 times per month [14]. We applied these criteria to our data except for symptom frequency as this information was not available in the GP notes. The second used a symptoms index (Index 2) that we developed in a previous study [12]. Women are considered positive on Index 2 if they have any one of pelvic abdominal pain or discomfort; loss of appetite; increase in abdominal size; able to feel a lump in the abdomen; or vaginal discharge. We also considered women who had two or more Index 2 symptoms given that the positive predictive value of individual symptoms is low [17].

Targeted ovarian cancer ‘screening’ is unlikely to be triggered by longstanding symptoms (unless there is worsening or change). Therefore, we also estimated the proportion of women who presented with a NICE guideline symptom that was new in the previous 12 months. The time period covered by downloading the last 10 consultation entries varied between women (median 6.0 months inter-quartile range 2.8-11.5). To use all available data for each woman we used the Kaplan-Meier method to estimate the proportion of women with a NICE guideline symptom in the reference week who had presented previously with that symptom as a function of time. From this we estimated the proportion who had at least one NICE guideline symptom for which they had not presented in the previous year (ie ‘new’). We multiplied this by the proportion of registered women aged 45–74 who presented with a NICE symptom to calculate the proportion (p) who would present during the week with a NICE guideline symptom that was new in the last year. The Delta method was used to calculate the variance to obtain confidence intervals for our estimate of NICE guideline symptoms new in the last year (taking into account the Kaplan-Meier estimate).

We calculated the proportion of women aged 45–74 who presented during the reference week with symptoms with 95% confidence intervals (for proportions) for women with:

We approximate the proportion of all women aged 45–74 that would be offered testing over one year according to different symptom categories using the equation:

According to our estimates, about a quarter of women aged 45–74 presenting to GP primary care would be offered symptoms-based screening for a NICE guideline symptom each year, however, this reduced to about a fifth when only ‘new’ symptoms were considered.

Despite having diverse locations, the four GP practices studied showed comparable proportions of women aged 45–74 who presented during the reference week. By contrast, the proportion of women in the target age range who presented with at least one ovarian cancer symptom relative to list size differed widely across surgeries.

A key strength of the study is that we collected data from a mixture of urban and non-urban practices to obtain a representative sample. In addition, having the last 10 consultations entries before the date of download provided information on symptom development. We used several testing scenarios to estimate the proportion of women who would be offered symptoms-based ovarian cancer screening, including an assessment of whether symptoms were new in the last year (testing is unlikely to be offered for longstanding symptoms). We used an age range that is most likely to be chosen for symptoms-based ovarian cancer screening; ovarian cancer incidence is much higher in postmenopausal women aged ≥45 and screening is not usually offered to those aged over 74. Finally, data were collected before initiatives to raise symptom awareness and encourage early presentation were commonplace (e.g. National Awareness and Early Diagnosis Initiative - NAEDI).

The main limitations are that the study was small and only records from women consulting during a single week were examined. This period is too short to take seasonal variations in consultations (e.g. flu season) into account. Additionally, we only studied four GP practices and these were seen to be heterogeneous. Another drawback is our inability to accurately assess other factors which would inform the decision to offer screening to women, such as symptom frequency, severity and persistence. In particular this could have led to overestimates of the proportion that would be offered symptoms-based screening.

Estimates from a recent UK study [18] were smaller than ours. The authors used logistic regression to derive three different weighted ovarian cancer scores were derived using GP medical record data from 212 women with ovarian cancer and 1060 age-matched controls. Seven symptoms were included in each of the scores: bloating, urinary frequency, rectal bleeding, postmenopausal bleeding, loss of appetite, abdominal pain and abdominal distension. The main score had a specificity of 91.3% which is equivalent to 8.7% of women without ovarian cancer testing positive on the score over one year.

A prospective study randomised GP practices to being able to refer (or not refer) women aged ≥45 with any one of a list of 19 symptoms potentially related to ovarian cancer for immediate CA125 testing and transvaginal ultrasound [19]. Compliance was poor and only 39 out of 79 practices made referrals during the study period. Referral patterns were also extremely variable between GPs. Over the recruitment period that spanned almost 2½ years, only 317 women were referred.

Based on our findings, symptoms-based screening would result in roughly the same amount of screening as two yearly population screening for any ovarian cancer symptom, four yearly population screening for any NICE guideline symptom and eight yearly population screening for any ‘new’ NICE guideline symptom. One drawback of symptoms-based screening is that it is necessary to wait for symptoms to develop and for women to present. Lower socioeconomic status is associated with delayed presentation for some cancers [20], therefore, symptoms-based screening could lead to an increase in the social inequalities that already exist in ovarian cancer [21].

Nevertheless, current referral criteria for rapid access gynaecological-oncology clinics (relevant to ovarian cancer) only include two symptoms: suspicious pelvic mass and postmenopausal bleeding (if not on hormone replacement therapy). Symptoms-based screening could be a useful tool when other typical ovarian cancer symptoms are present but the index of suspicion is not sufficient to prompt referral or more expensive investigations. This could be particularly appealing given that the effectiveness of population-based ovarian cancer screening remains unclear. UKCTOCS (United Kingdom Collaborative Trial of Ovarian Cancer Screening) is the largest randomised trial of ovarian screening to date. Around 200,000 women from the general population were randomised to either screening with transvaginal ultrasound (TVS) or multimodal screening with CA125 and TVS or no screening. Data from the prevalent screening round found that sensitivity for primary invasive epithelial ovarian/tubal cancer was 89.5% and specificity was 99.8% in the multimodal arm [22]. Although these data are encouraging, the final picture will need to be balanced against unnecessary surgeries, complications and most importantly, needs to be able to demonstrate a mortality benefit. Mortality data are due in 2015 and are expected to provide major insights into ovarian screening.

Although our estimates for women who would require symptoms-based testing are relatively large, the extrapolations over one year are somewhat crude. Furthermore, any testing threshold for symptoms-based screening is likely to include a criterion for symptom persistence which was unaccounted for in this analysis. The NICE guidelines [14] and the symptom index produced by Goff et al. [23] stipulate that symptoms should occur more than 12 times per month Applying similar criteria should further reduce the number of women who require testing, as would the use of a score or similar. In this study, only 3.6% (95% CI 1.0%-8.9%) of women aged 45–74 would have had testing in one year if the requirement was for two or more Index 2 symptoms. Additionally, testing would not be offered to women who have already been recently tested (say the last 6–12 months) which would also reduce testing numbers.