Refine baseline screening coverage and frequency

[carajbro]

Reaching out to Sinead and other experts in South Africa due to discrepancies in reported data.

[carajbro]

Summary of data on cervical cancer screening by Darcy: The District Health Barometer annual reports present data from the National Health Laboratory Service on the number of cervical smears taken in women aged 30 and older. To estimate the coverage of screening, they divide this by 10% of the size of the female population aged 30 or older, under the assumption that women screen every 10 years in line with guidelines. It does not appear that they account for the fact that women living with HIV are recommended to start screening earlier and more frequently.

• Based on these data, an estimated 79.9% of women in KwaZulu-Natal screen according to guidelines, and 61.2% of women in all of South Africa do.
• However, using this method, some districts are estimated to have >100% coverage. This suggests that there are some data quality issues. There may also be some groups of women who screen more frequently (including WLHIV), such that the denominator is not accurate.

A 2016 paper by Makura et al. analyzed 2013/2014 NHLS data using the same approach as the DHB reports and estimated the median coverage to be 33% (IQR 26-42% across districts). This is less than the DHB report from the same year, which estimated coverage to be 54.1%. As far as I can tell, these analyses used the same data and a similar approach, although the Makura paper may have taken differential screening patterns for WLHIV into account in defining the denominator for the general population estimate. The explanation they give for the discrepancy is that there may be under-reporting of women attending facilities or that the facilities are under-performing, but since the data sources are the same as far as I can tell I’m not sure I understand.

• This paper also reported on the pap smear adequacy, which was 47% overall. The denominator for this was smears that were “satisfactory for examination,” so the proportion of all smears that they included in the calculation of coverage that were adequate would be even lower.

The 2016 South Africa DHS reports on self-reported receipt of a Pap. Nationwide, 37% of South African women aged 15 and older ever had a Pap, 88% of whom had the test in the past 10 years. Among women aged 30-59, 52% ever had a Pap, 92% of whom had one in the past 10 years. In KZN, 32% of women aged 15 and older ever had Pap (88% in the past 10 years), and 48% of women aged 30-59 ever having a Pap. These data are self-reported, so they may be biased, but this suggests that the DHB data may be overestimates.

A 2017 paper by Jordaan et al. discusses the history, current status, and challenges to cancer screening in South Africa. They present NHLS data and highlight that implementation of the screening guidelines has been challenging. In particular, they point out that the data do not show that guidelines for WLHIV are being implemented, as one would expect to see increases in the number of screens conducted coinciding with changes in guidelines for screening of women living with HIV. They note that a shortage of healthcare personnel and equipment are barriers to screening, and many women are put on wait lists for triage and treatment that can last up to 18 months.

From this, we need to decide how to implement screening in our model up to 2020, both in terms of the ages at which women are screened and the coverage at those ages. An additional challenge with the model, as we discussed, is that our current model structure does not allow us to implement repeat screening for the same people. If we assume screening every 10 years in recent years at, say, 60% coverage, this will in effect correspond to screening virtually all women at least once rather than serially screening 60% of the population three times.

We separately implement parameters to control the proportion of women screened who have a true positive diagnosis, the proportion who are triaged, and the percent treated. Our current assumptions are below:

• Cytology is 57% sensitive for detection of CIN2+
• 72% of those diagnosed return for colposcopy and are correctly diagnosed
• 51% of those with CIN2/3 are treated, and treatment is assumed to be 77% effective for HIV-positive women and 91% effective for HIV-negative
• 40% of women with cervical cancer are treated with 100% efficacy.

So the questions are:

• What assumptions about screening coverage and intervals seem realistic? In 2020? In 2010?
• Are WLHIV screened more frequently, as recommended in the guidelines?
• What proportion of women who screen positive are treated?

[darcyrao]

Notes from call with Admire Chikandiwa on July 21, 2020

[carajbro]

Issue of how to apply screening in a compartmental model with 5-year age groups (from Darcy's email to the WHO modeling group on 9/30/2020)

Let’s say the scenario we are implementing is that 50% of women are screened at age 35. Below is a simple representation of what that would look like with a single-year model, which approximates the “reality” we are trying to represent. 50% of women in the 35-year age bin are screened. If we then look at women older than this and were to ask what proportion were screened at the targeted age, we would find that 50% of them were, as expected.

Let’s assume that we have a small population with 100 women at each age. If we put this in numbers, this corresponds to 50 women getting screened at age 35. To implement this in a model with 5-year age groups, one option, which is what we originally tried, is to preserve the number of women screened by applying the coverage to 1/5 of the 35-39-year age group. But, since we have a memoryless process with compartmental models, the 50 women screened get distributed throughout the 5-year period. So essentially, this scenario is the same as saying that 10% of all women in this age group get screened. Below we see how this looks in the year screening is first implemented.

In the next year, 1/5 of the women who were in the 35-39 year age group will have aged into the 40-44-year age group. On average, 10% of these women will have been screened in the 35-49 year age group, which is 20% lower than would happen in the scenario we are trying to represent.

At the same time, 100 women from the 30-34 year age group will age in to the 35-39 year age group. None of these women will have been screened, so the model will screen enough so that the screening coverage in the 35-39-year bin remains at 0.5*0.2 = 0.1. This results in only 10 women being newly screened.

Now, the way that we have decided to implement this is by applying coverage to the entire age band. This is in effect modifying the scenario slightly to say that 50% of women aged 35-39 will be screened. Women are screened only once in this age band. 50% of new cohorts aging into this age band will be screened, and if we look at women aged 40-44, 50% of them will have been screened between the ages of 35-39. This scenario not only more accurately represents the scenario we are trying to represent, but it is also more realistic — it is unlikely that a program would actually target only women aged 35 years old.

[carajbro]

Alternative approach to consider in the future:

• Could screen a certain number of unscreened women each year (at a constant rate) rather than bringing up coverage to a certain level.
• This method still would not specify re-screening in certain women