Distance as Explanatory Factor for Sexual Health Centre Utilisation in an Infrastructure-Rich Urban Area in the Netherlands: A Population-Based Multilevel Study

utilise the SHC, also including the utilisation rate per 1,000 residents with 95% condence intervals (CI) for the study population and the STI positivity rate with 95% CI among SHC users. The STI positivity rate is the percentage of SHC users with one or more STI diagnoses (chlamydia, gonorrhoea, infectious syphilis, HIV or infectious hepatitis B), and gives insight into area specic high-risk STI population subgroups. For each PC area we geographically presented the degree of urbanisation, ethnic diversity, and the utilisation rate per 1,000 residents. We also plotted distance against utilisation rate per PC area.


Conclusion
Living further away from the sole SHC in the greater Rotterdam area decreases utilisation. This provides evidence for local policy to enhance STI testing, for example by offering STI testing services closer to the population.

Background
Early diagnosis and adequate treatment are essential in controlling sexually transmitted infections (STI), including HIV. Many infected people remain untested and untreated. Easier access to testing services and subsequent treatment can improve health outcomes of people with STI, and reducing the risk of transmission (1).
In the Netherlands, STI tests and treatment are mainly provided by general practitioners (GPs) and sexual health centres (SHCs). The SHC is restricted to those considered high risk for STI and those who need sexual health advice the most (e.g. being noti ed for an STI, having STI symptoms, having a non-Western migratory background, aged below 25), which is assessed through triage (2). Those who do not meet at least one of the SHC triage criteria are advised to visit a GP. In contrast to the GP, SHCs are funded by the government, making it possible to offer tests and treatment free of charge (2). The GP will not charge for an STI consultation, however STI tests are only free of charge if the deductible excess of the health insurance is paid (at least Euro 385) (3).
Financial cost is one of the barriers for healthcare utilisation and testing (4)(5)(6). As the SHC service is completely free of charge, nancial barriers should not play a major role in approaching an SHC. There are several other barriers (e.g. service access, perceived needs, social-cultural factors) that undermine healthcare utilisation and, hence, testing (4)(5)(6)(7)(8)(9)(10)(11)(12). This study is interested in how geographical proximity acts as a barrier to SHC utilisation. Various studies have identi ed geographical proximity as an important structural factor to explain inequalities in geographically accessibility (13)(14)(15)(16). Utilisation of a healthcare service, as a proxy for accessibility, appears to decrease with an increasing travel time or distance (13)(14)(15)(16). We could not nd any quantitative studies investigating the effect of distance on SHC utilisation in western countries.
Based on the hypothesis that larger travel distance is inversely associated with SHC utilisation, we conducted a population-based study with the aim to determine a possible association between SHC utilisation and travel distance in the greater Rotterdam area. Con rmation of the hypothesis would provide local policy makers with evidence to enhance the (geographical) accessibility to SHC services and thereby increase STI testing and treatment rates.

Study area and SHC location
This study focuses on the sole SHC located in the city of Rotterdam. This SHC is run by the municipal public health service. The greater Rotterdam area, consisting of the city of Rotterdam and 14 neighbouring municipalities, harbours 1.3 million residents, half of them living in Rotterdam. The river Maas divides both the greater Rotterdam area and the city of Rotterdam into a northern and a southern part. The SHC is situated in the northern part, very close to a bridge connecting the northern and southern part, and with both a subway and tram stop in front of the building.

Data sources and study population
The study population consists of all residents aged 15 to 45 years in the greater Rotterdam area, obtained from the Dutch population registry (Statistics Netherlands). Each person in this registry has a unique citizen service number (BSN). Due to privacy legislation, the BSN is not collected during SHC consultations. Therefore, we matched each SHC consultation record to an arbitrary, unique resident in the population registry by year of consultation, year of birth, sex, grouped migratory background and fourdigit postal code (PC). From the SHC consultation database, we only selected the rst SHC consultation of each attendee that met the following criteria: a heterosexual man or woman living in the greater Rotterdam area, aged 15 to 45 years, and visiting the SHC in Rotterdam for an STI test. We made this choice because most of the general population is heterosexual and because the proportion and residential distribution of men who have sex with men (MSM) in the general population is unknown. In addition, more than 95% of all SHC heterosexual attendees belong to the age group 15-45 years. We used population registry and SHC consultation data from 2015, 2016 and 2017. Additional data from Statistics Netherlands (degree of urbanisation) and the Netherlands Institute for Social Research (socioeconomic status [SES]) were also linked to the dataset by PC.

Outcome variable
The main outcome of interest was access to the SHC in Rotterdam, operationalised as SHC utilisation. Only residents that match with the SHC consultation database are assumed to have utilised the SHC.

Determinants
Both determinants at individual and PC level are considered (Supplementary Table 1). The individual determinants included sex, age and grouped migratory background. The main determinant of interest on the PC level was travel distance to the SHC. Other PC level determinants included degree of urbanisation, SES, ethnic diversity, and living in the northern or southern part of the greater Rotterdam area. Since travel distance (straight-line and road-network) and travel time were highly correlated (r 2 >0.9), straight-line travel distance between the centroid of the PC area and SHC address was used as proxy for travel distance. Ethnic diversity is measured by the Her ndahl-Hirschman Index. The index can be interpreted as the probability that two randomly selected individuals from the same PC area belong to different migratory background groups. We included living in northern or southern part of the area as determinant because we hypothesized that the river Maas, dividing Rotterdam into northern and southern areas, may serve as natural barrier for SHC utilisation.

Statistical analyses
Potential selection bias was assessed by comparing selected consultations for SHC attendees that match the population registry to consultations without match. Only records with complete data for all determinants were included in the analysis. Descriptive analysis was performed to describe the study population and those who utilise the SHC, also including the utilisation rate per 1,000 residents with 95% con dence intervals (CI) for the study population and the STI positivity rate with 95% CI among SHC users. The STI positivity rate is the percentage of SHC users with one or more STI diagnoses (chlamydia, gonorrhoea, infectious syphilis, HIV or infectious hepatitis B), and gives insight into area speci c high-risk STI population subgroups. For each PC area we geographically presented the degree of urbanisation, ethnic diversity, and the utilisation rate per 1,000 residents. We also plotted distance against utilisation rate per PC area.
Since the hierarchical structure of our data, residents located within 183 PC areas in 15 municipalities, we conducted multilevel logistic regression analyses. The top level of the hierarchy (municipality) was not modelled, since the small number of municipalities (n=15) produced unreliable estimates, and policy implications would most likely target PC areas. First, a null model (Model 0), only adjusting for year, was constructed to see whether PC variance was signi cant. Second, univariable models were computed.  visiting the SHC, and 0.5 suggests that the model is equivalent to random guessing. For each model, we also calculated the proportional change in the variance (PCV), with the null model as reference, and the median odds ratio (MOR). The PCV indicates the explained PC area variance by the determinants in the model. The MOR is used for quantifying the magnitude of the effect of clustering, and can be interpreted as the median increase in likelihood of visiting an SHC when moving a person from the PC area with lowest probability of visiting the SHC to a PC with a higher probability of visiting the SHC.
Before the models were constructed, we checked for bivariate Pearson correlation between variables. The correlation ranged from 0.0 to 0.7. Multicollinearity was de ned by a variance in ation factor (VIF ≥10).
No determinants were excluded based on correlation; all variables had a VIF<5.
SHC triage policy does affect the utilisation rate for certain groups for whom triage is performed (aged below 25 and/or having a non-Western migratory background have higher 'priority'). We, therefore, also performed the same analyses separately for residents aged below 25 and for non-Western migratory background. It was not possible to perform a multilevel analysis for a combined strati cation of age and migratory background, since the number of SHC visitors became too small to reliably estimate differences between PC areas.
All statistical analyses were conducted using SPSS (version 26). P-values were 2-sided and P<0.05 was considered statistically signi cant.

Data selection and matching
For each study year, we included over a half million residents, with 1,582,017 records in total. Of the 19,460 SHC consultations that ful l the study inclusion criteria, 220 (1.1%) records could not be matched to the population registry. There were no signi cant differences in individual determinants and triage criteria (e.g. being noti ed, having STI symptoms) between the matched and non-matched group. Only records with complete data were included in the analysis. In total, 646 records (0.04%) had to be excluded due to unavailability of SES information. This left 1,581,371 resident records with 19,237 SHC consultation record matches for analysis (Table 1).

Study area and study population
Based on the utilisation rate per 1,000 residents, SHC visitors were more often women, below 25 years, non-Western, and living in highly urbanised or low SES areas ( Table 1). The straight-line distance from PC area to the SHC ranged from 0.6 km to 41.2 km. In general, the SHC utilisation rate decreased with increasing distance to the SHC ( Figure 1D and Table 1). PC areas relatively close to the SHC are also the areas with a higher degree of urbanisation and a more ethnically diverse population ( Figure 1A and 1B).
The SHC utilisation rate between PC areas ranged from 1.13 to 48.76 per 1,000 residents ( Figure 1D).
The overall positivity rate was 21.1% (95% CI: 20.5%-21.7%) among the SHC visitors. In general, the positivity rates for the various subgroups differed little from this overall positivity rate. The positivity rate was lowest for visitors aged 25 years or older (16.8%) and highest for Cape Verdean visitors which also had the highest utilisation rate (31.3%).

Multilevel models for SHC utilisation
Multilevel logistic models for SHC utilisation are presented in Table 2 Table 2). After adjusting for travel distance and individual-level determinants (Model 1), the PC variance decreased by 77.5% to 0.15 compared to the null model, as shown in Table 2. Adding other PC area variables to the model (Model 2) explained 87.0% of the PC variance, leaving a MOR of 1.33. In other words, if a resident moved to another PC area with a higher probability of utilising the SHC, the median increase in their odds of utilising the SHC would be 1.3 fold (MOR = 1.33).
In Model 2, which adjusts for individual and PC determinants, living closer to the SHC was associated with SHC utilisation ( Table 2). Each kilometre increase was associated with 5% decrease (OR: 0.95; 95% CI: 0.94-0.96) in the odds of utilising the SHC. This means that a person has a 20% lower odds of utilising the SHC (OR: 0.81) when residing at 8.0 kilometres (75 th percentile of distance) compared to 4.0 kilometres from the SHC (25 th percentile). The ORs of the individual-level variables in Model 2 were similar to the ORs observed in Model 1 ( Table 2).
Each variable included in Model 2 decreased PC area variance, ranging from -0.6% for sex to -31.3% for age and -32.8% for travel distance (Table 3). Travel distance and ethnic diversity appeared to be the most important PC determinants in PC area variance decrease in Model 2.
Strati ed multilevel models for SHC utilisation The same analyses were performed for residents aged under 25 (Supplementary Table 3) and for residents with a non-Western migratory background (Supplementary Table 4). Among residents aged below 25, similar results were observed to the overall results (Table 2); the OR for distance was 0.95 (95% CI: 0.94-0.96) in the nal model, and the VPC and MOR had a similar pattern with a nal MOR of 1.33.
The results for non-Western residents differed from the total population and the residents aged below 25.
Univariably (data not shown), distance was statistically signi cantly associated with SHC utilisation (OR: 0.94; 95% CI: 0.93-0.95), which was not the case in the nal model for non-Western residents (OR: 0.99; 95% CI: 0.99-1.00). Only age and migratory background were statistically signi cant associated in the nal model. The PC variance was fully explained for both Model 1 and 2 (PCV=100%), with a corresponding MOR of 1.
Travel distance accounted univariably for the largest decrease in PC variance for both residents aged below 25 and non-Western residents (data not shown). The MOR for the univariable model with travel distance was 1.49 for residents aged below 25 and 1.34 for non-Western residents.

Model performance
The AUC improved from 0.505 in the null model to 0.819 in nal Model 2, re ecting a good discriminative ability of SHC utilisation ( Table 2). As shown in Table 3, age had the largest added value in model performance, since the AUC decreased most when age was removed from Model 2 (AUC=0.714). Distance was the second-best determinant in model performance (AUC=0.802), together with individual migratory background and postal level ethnic diversity (for both AUC=0.803). The discriminative ability of both the nal model among residents aged below 25 and among non-Western residents was less compared to the overall model, with respectively an AUC of 0.733 and an AUC of 0.775.

Discussion
Our analysis con rmed the hypothesis that larger travel distance is inversely associated with SHC utilisation in the greater Rotterdam area. The distance decline is independent of age and migratory background. Of all variables included in the nal model, travel distance accounted for the largest decrease in PC area variance.
The results of our study are consistent with other literature (13)(14)(15)(16). However, these studies are not speci cally for SHC utilisation and many of these studies are not in Western infrastructure-rich urban areas, like the Netherlands. We found the same distance effect for people aged below 25, but not for people with a non-Western migratory background. Possible explanations are related to the provider, the client and area (demographic) characteristics. Triage is probably the most important explanation on provider side, because prioritisation makes SHC consultation generally more accessible for people with a migratory background. Residential location is not an SHC triage criterion, however migratory status is prioritised above the under 25 years old criterium since migrants' STI positivity is generally higher (2,17,18). Difference in utilisation seems not to be affected by other triage criteria than age and migratory background; no difference was observed with other prioritised triage criteria, i.e. being noti ed or having symptoms.
An explanation on client side is self-selection or (non) familiarity, to explain the difference in distance effect. Those living further may be more critical on their perceived STI risk, since it takes more effort to visit the SHC. From the literature it is known that a higher risk perception is positively associated with STI testing (11,(19)(20)(21). It may be that the anonymous consultation and free of charge services offered by the SHC are more important for migrants to counterbalance the distance (22,23). Previous research showed that more distant healthcare facilities may actually be preferred for stigmatized health conditions (24,25). It is known that migrants perceive more shame and stigma related to STIs (5,26). Also, perceived issues with con dentiality and privacy at the GP may play a role in choosing anonymous STI testing at the SHC (4, 27).
Another explanation for the difference in distance effect could be a difference in sociodemographic distribution among PC areas or on non-measured determinants. Migrants may reside generally further away from the SHC or at places with good public transport access compared to youngsters, affecting the UR. From additional analysis we could conclude that migratory groups with a high utilisation rate in our study (Antillean, Surinamese or Sub-Sahara African), reside all over the region without clear 'migrant neighbourhoods'. Turks and Moroccans tend to reside slightly more remote from the SHC and show more area clustering.
We were able to explain a substantial proportion of the variance between PC areas. In the overall model, the PC variance in SHC utilisation decreased with 87.0%. Distance explained most decrease in PC variance. Distance had also the second-best added value (together with individual migratory background and postal level ethnic diversity) in model performance, after age. Since age and migratory background are already part of the triage policy, this nding strongly suggests introducing policy measures that decrease the access inequality between areas caused by distance, for example by using a mobile clinic, an additional location, or community-based testing in more remote areas. Despite increasing the access by lowering the physical distance, a MOR of 1.33 in the nal model still indicates a substantial difference between PC areas in SHC utilisation even when other individual and area determinants are similar. This implies that we did not model all (area) determinants explaining geographical differences in utilisation.

Strengths and limitations
Strengths of this study are rstly that this appears to be the rst large-scale study linking SHC consultation data to population data to investigate SHC utilisation in high-income areas. Secondly, we used multiple data sources for the fullest possible set of determinants. Thirdly, our multilevel approach allows the simultaneous examination of factors at different levels, in our case individual and PC area. Therefore, we were able to demonstrate the importance of area level determinants, which is often lacking in other studies. Finally, we carefully considered our distance measure. We calculated multiple measures for proximity, which were all highly correlated (r 2 > 0.9). Other studies also found that straight-line distance is an adequate proxy for road network distance and travel time in more urban areas (28)(29)(30)(31).
The major limitation of the study is that we are not able to quantify the clinical signi cance of lower utilisation rate among more remote areas from the SHC. If residents in these areas have a lower STI risk and are not visiting the SHC, or instead visiting the GP, this is less severe than high STI risk residents not visiting both SHC and GP. Similarly, we do not know if residents close to the SHC are utilising it as needed. Based on the STI positivity by distance, it can be argued that persons with a high risk nd their way to the SHC by mechanisms such as triage and self-selection, as explained earlier. On the other hand, the STI positivity for those living further away is not lower, while the SHC utilisation rate is. Since it is very unlikely that the distance effect only applies to low STI risk individuals, the SHC does not reach everyone that should be reached. To better interpret these results, and to develop an optimal strategy for local STI testing services, further research is needed to address the role of the GP. Another limitation is that we are not able to completely correct for triage effect. We have no information on triage criteria for all residents, or more speci cally, for those who are rejected for an SHC consultation based on triage or limited consultation availability. Insight in the rejected individuals would give more insight in the 'real' SHC accessibility and missed opportunities. We know that almost everyone who attempts to make a consultation at the SHC in Rotterdam has at least one triage criterion (unpublished data). From this same work, it was also concluded that a signi cant proportion high-risk people were refused due to limited consultation availability. Finally, we assigned the same distance to the SHC for all residents in one PC area. A more individual calculation of distance was not possible since anonymous consultation data only contain four-digit PCs. Nevertheless, several studies have shown that centroid distance is an acceptable proxy measure (32,33).

Conclusion
Our ndings are relevant for local policy as we demonstrate that distance is a signi cant barrier for STI testing at the sole SHC in this infrastructure-rich urban area. Minimising travel distance, e.g. by using mobile clinics or community-based testing, in more remote areas, could be a strategy to reduce area differences in STI testing. Different strategies may be considered for different subgroups. No ethical approval was needed under prevailing laws in the Netherlands as this study is a retrospective observational study using anonymous data only (as stated by the National Central Committee for Human Studies: www.ccmo.nl and in the conduct of good behaviour in research www.federa.org).

Consent for publication
Not applicable.

Availability of data and material
The data that support the ndings of this study are available from the corresponding author on reasonable request.

Declaration of competing interests
The authors declare that they have no competing interests.

Funding
This research did not receive any speci c grant from funding agencies in the public, commercial, or notfor-pro t sectors.
Authors' contributions DT initiated the study, analysed and interpreted the data and drafted the manuscript. HG and BM initiated the study, helped interpreting the data and revised the manuscript. DN advised in the statically analysis, its interpretation, and revised the manuscript. JHR helped interpreting the data and revised the manuscript. All authors read and approved the nal manuscript.    Table 2) as reference. Figure 1 Degree of urbanisationa, ethnic diversityb, SHC utilisationc and SHC utilisation by distance to SHCd Abbreviations: km, kilometre; SHC, sexual health centre. The basemap ( Figure 1A, 1B, 1C) is created with publicly available data from Statistics Netherlands. The data presented in these maps are based on publicly available data ( Figure 1A) or data generated in this study ( Figure 1B  concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.