Assortative sexual mixing in a heterosexual clinic population - a limiting factor in HIV spread?

Objective: To assess the degree of sexual mixing in a sexually transmitted disease clinic population stratified by country of birth.

Design: Prospective linked HIV serosurvey incorporating demographic and sexual risk data gathered by a doctor-administered questionnaire.

Setting: The Department of Genitourinary Medicine at St Thomas' Hospital, London, UK.

Subjects: Fifteen thousand eight hundred and seventy-eight heterosexuals who attended between April 1992 and February 1995.

Main outcome measure: The degree of assortative (like-with-like) mixing, after stratification of the population by country of birth, of index patients, their parents and their sexual partners.

Results: Sexual mixing in this population of sexually transmitted disease clinic attenders is highly assortative when the CoB of parents (family origin) of index patients is taken into account.

Conclusions: Our findings help to explain the low spread of heterosexual HIV infection in the UK to date, and may help future projections and health targeting of those at risk. This model can be applied to other mixed populations.

Introduction

Since 1988 [¹], projections of AIDS cases among heterosexuals in England and Wales have been successively lowered [^2-4]. By December 1996, 5433 HIV infections following sex between men and women had been reported [⁵]. Excluding 633 with 'high risk partners' and 321 'undetermined', 4007 (89.5%) of 4479 heterosexually exposed individuals had acquired their HIV infection overseas, the majority of these, 3394 (84.7%), in Africa.

Uncertainty remains about an epidemic dominated by infection acquired overseas because the extent of sexual mixing among heterosexual subgroups is unknown [³]. This uncertainty continues to frustrate projections [⁴]. The Department of Genitourinary Medicine at St Thomas' Hospital serves an inner-London population with diverse national origins. In common with the rest of the UK, we have found that the majority of HIV infections, following sex between men and women, have been imported. Since 1988 we have conducted a linked HIV serosurvey using a doctor-administered questionnaire that records a variety of data, including the country of birth of index patients, their parents and their sexual partners.

Some models of HIV transmission [^6,7] have concentrated on mixing between groups stratified by rates of partner change, but other parameters may be informative [⁶]. We have stratified partnerships amongst our attenders by CoB. If sexual mixing in the UK were proportionate (random), one would expect partnerships involving overseas-born index patients to be predominantly with UK-born persons, since the latter make up the majority of both the clinic population under review and the general population.

Methods and results

As part of a long-term HIV serosurvey in the Department of Genitourinary Medicine, a questionnaire was administered to all patients who were also undergoing a blood test (usually for serological tests for syphilis) before inviting them to take an HIV test. The questionnaire was filled in by the doctor who saw the patient and included details of sexual activity during the previous 12 months in and outside of the UK. Demographic data included the patient's, their parents', and their partner(s)' countries of birth.

Table 1 gives the country of birth of 15 878 heterosexual men and women attending between April 1992 and February 1995. The UK accounts for 10 919 (68.8%), the remainder including 1498 (9.4%) born in the West Indies (WI), 1130 (7.1%) in Europe (excluding UK) and 1129 (7.1%) in (sub-Saharan) Africa. This preliminary analysis is restricted to these four geographical groups (92.4% of total and 25 678 partnerships) and only concerns sexual activity in the UK, in the 12 months preceding interview.

Table 1

There was no limit to the number of partners that could be recorded but a maximum of three fields for countries of birth of sexual partners of each patient in the UK was available. Data potentially lost because of this restriction related to 93 of 7451 men (1.2%) and 57 of 8427 women (0.7%), those who named three partners countries of birth and reported more than three partners. In 13% of male and 9.5% of female partnerships, there was a mismatch between numbers of partners' and countries (for instance, two countries and three partners). In these cases, we calculated a minimum and maximum possible number of partners for each country of birth. Since in no case did using either maxima or minima make any significant difference to the results (available but not shown), we have taken the mid-point between minimum and maximum for the purposes of calculation. Table 2 shows the partnership data, each cell being made up of this median added to the number of known partners.

Table 2

Partnerships were next stratified by family origin, taking into account parents' country of birth. For example, looking at partnerships involving UK-born index women with men born in Europe, 75 (19%) of 390 index women (Table 2) had at least one parent born in Europe. Adjustment for family origin moved these 75 partnerships from the cell UK/Europe men in Table 2 to the cell Europe/Europe men in Table 3. Equivalent calculations for partnerships of UK-born index women with men born in Africa and the WI showed 124 (32%) of 385 to have African parents and 696 (68%) of 1019 to have West Indian parents, respectively.

Table 3

Where a patient born, for example, in the UK had mixed parentage (for instance, UK and another) and a partner from the non-UK parent's region, this partnership was included in the adjustments detailed above because of the link (albeit less strong than with concordant parents) in family origin. Such mixed parental adjustments contributed 432 (24.6%) of 1834 movements between Table 2 and Table 3 and were predominantly between UK and Europe (301 out of 432, 69.7%). For Africa and WI, the number of mixed parents ranged from six (4.5%) of 133 partnerships with African women to 92 (13.2%) of 696 partnerships with West Indian men. Removing these 432 partnerships entirely made only a small difference to either Table. In Table 2, 11 out of 24 cells showed changes in percentages of more than 1%, the largest increase being in WI/WI women (+3.3%) and the largest decrease being UK/WI women (-3.6%). In Table 3, six out of 24 cells showed changes in percentages of more than 1%, the largest increase being in UK/Europe women (+2.0%) and the largest decrease being Europe/Europe women (-2.1%).

When adjusting for parents' country of birth, we have taken account of patients born in countries other than the four regions under study. These 'others' contributed 168 to the UK (i.e. those born in other countries but of UK-born parents 'gave' 168 partnerships), six to Europe, and none to Africa or WI.

We have no data on the country of birth of partners' parents, but one might hypothesise that partners would have the same proportional mix of parents as index cases. Thus, we know that 696 (68.3%) of 1019 UK-born index women who had sex with West Indian-born male partners, had West Indian parents. Applying the same proportion to the 1191 UK-born women partners who had sex with West Indian-born index males, 813 (1191 × 0.683) would also have West Indian-born parents. Applying these relevant known proportions to partners in Table 3, generates Table 4 (i.e. 813 move from WI/UK women to WI/WI women, etc.). We have looked at 400 linked partnerships (where both attended this clinic) and found that, of 22 West Indian-born men with UK-born partners, 18 (82%) of the latter were of West Indian family origin. Although the numbers are small, they support the suggestion that Table 4 most nearly describes the degree of mixing according to family origins.

Table 4

'Expected' values for each cell in Tables 2 and 3 have been calculated by multiplying the proportional contribution of each column to the total, by the gender row totals (results not shown). Figure 1 displays observed and expected values for Tables 2, 3 and 4 expressed as percentages. All the dissimilar pairings (say, European male with UK female) show lower observed than expected values in all tables even when maxima or minima are used (results not shown), and the converse is true for like-with-like pairings.

Fig. 1

Using the Poisson distribution to compare expected with observed readings, overlapping 99% confidence limits for λ were found for Europeans with Africans, and for West Indian women with African men in Table 3. For all like-with-like pairings in all tables, observed values were greater than expected, and none of the 99% confidence limits for λ overlapped.

The use of the available partnerships to generate expected (random) values deserves comment. There is no defined population that is equivalent to those attending our central London, open-access clinic. It will be apparent from Fig. 1, however, that, whichever population is chosen for comparison (for instance, all of UK, the south of England, inner London, etc.), the expected distribution of partnerships in any chart will always be identical for each CoB and will therefore always be different from the observed values which do not show random (proportionate) mixing.

Mean numbers of partners in the past year were higher for men: UK-born 2.35 (variance 6.84); WI-born 2.24 (9.05); European-born 2.07 (8.04) and African-born 1.76 (6.43); than for women: UK-born 1.61 (4.96); WI-born 1.24 (2.23); European-born 1.66 (5.29) and African-born 1.28 (2.43).

Discussion

Potential sources of bias in this study relate to the accuracy of self-reported sexual activity; for example, in common with other studies [⁸], men report more sexual partners than women. The serosurvey questionnaire forms were not filled in for a number of patients, which was estimated at 15%, with the only discernible pattern of non-completion being doctor-related rather than patient-related. Missing data for this analysis ranged from 1% for CoB of index patients to 3.5% for CoB of partners.

Stratification of partnership data by CoB demonstrates assortative mixing (Table 2). The results of further stratification by family origin, taking the CoB of the index patients' parents into account, are seen in Table 3. In all like-with-like pairings, the degree of assortativeness was increased in Table 3 when compared with Table 2. The hypothesis generating Table 4, which supposes that partners have the same proportional mix of family origin as index patients, when considering partnerships by country of birth, is intuitively reasonable and increases assortativeness still further.

While the population attending a sexually transmitted disease (STD) clinic is not representative of the general population, it includes those at greater risk of HIV infection. For example, numbers of partners in the past year are greater than those found in the general population [⁸].

Earlier data from this unit [⁹] indicated that HIV infections acquired overseas occurred predominantly in nationals of the countries involved, rather than in travellers originating in the UK. This finding was endorsed by data from centres outside [¹⁰] and inside London [^11-14]. Our HIV serosurvey findings continue to support this observation [^15-18] (and Barlow D et al., in preparation) and is implicit in the caveats in the two recent Day reports [^3,4]. Further support for a highly assortative pattern of sexual mixing in our clinic population is given by the pattern of gonorrhoea among our heterosexual patients which remains concentrated in sub-populations of our attenders [^19-23] and constitutes a local endemic [²⁴] that appears to be geographically and ethnically confined.

These findings help to explain the pattern of spread of HIV amongst heterosexuals within the UK to date and could assist in developing predictions in the future. Our clinic population includes a significant number born in the Caribbean (9.4%) and sub-Saharan Africa (7.1%) but few from the Indian sub-continent or South-East Asia where a rapid increase in HIV incidence has been reported. Similar analyses from other STD clinics inside and outside London, particularly with sub-populations from these two regions, may point to future trends in HIV infection. This approach may help to identify subgroups with a higher risk of infection in other parts of the world with mixed populations and, most importantly, will allow targeted health promotion for those at particular risk of infection.

Further analysis of our data will include stratification by age, sexual activity and ethnic group, and will look for any differences between those with single and multiple partnerships. We shall also examine data relating to sex outside the UK, including numbers of partners, their CoB and countries of travel. Analysis of a previous cohort [²⁵] indicated differences between sub-groups when travel to parts of the world with high HIV prevalence was considered.

Acknowledgements

We thank the patients who supplied, the doctors who gathered and the clerical staff who entered the data presented in this paper. We also thank D. Band and R.N. Thin for advice.