Late HIV diagnosis: An application of flexible Bayesian approaches to understand associated factors and time to diagnosis in Ireland, 2015–2024

Cian Dowling‐Cullen; Katie O'Brien; Kate O'Donnell; Mary Archibald; Phil Downes; David Field; Caroline Hurley; Fiona Lyons; Eavan Muldoon; Peter Barrett; Cathal Walsh; Derval Igoe

doi:10.1111/hiv.70311

. 2026 Sep 24;27(11):1881–1895. doi: 10.1111/hiv.70311

Late HIV diagnosis: An application of flexible Bayesian approaches to understand associated factors and time to diagnosis in Ireland, 2015–2024

Cian Dowling‐Cullen ¹, Katie O'Brien ², Kate O'Donnell ², Mary Archibald ², Phil Downes ², David Field ³, Caroline Hurley ⁴, Fiona Lyons ^4,^5,⁶, Eavan Muldoon ^7,^8,⁹, Peter Barrett ^10,^11,^12,^✉, Cathal Walsh ^2,⁵, Derval Igoe ¹

PMCID: PMC13725708 PMID: 42785996

Abstract

Objectives

Late HIV diagnosis is associated with increased morbidity, mortality and onward transmission. The proportion diagnosed late reflects the combined effects of testing delays, incidence and migration, complicating its interpretation. We analysed national HIV surveillance data for Ireland (2015–2024) to understand the factors associated with late HIV diagnosis and testing delays.

Methods

We estimated prevalence ratios (PRs) for late diagnosis among new diagnoses using Bayesian logistic regression with marginal standardization. We investigated testing delays by estimating infection times and modelling truncation‐adjusted delay distributions with Bayesian lognormal regression. We conducted several sensitivity analyses for the main findings.

Results

Among 1551 included new HIV diagnoses with available information, 684 (44.1%) were late. This proportion exceeded 50% in several subgroups, including people aged over 50 years, people whose probable mode of transmission was heterosexual contact, and people born in sub‐Saharan Africa. Increasing age (adjusted PR per decade 1.20; 95% credible interval [CrI] 1.15–1.26) and heterosexual transmission (adjusted PR for heterosexual men vs. sex between men 1.43; 95% CrI 1.23–1.67) were associated with late diagnosis in both univariable and multivariable models. For in‐Ireland infections, median time from infection to diagnosis was estimated at 1.94 years (95% CrI 1.36–3.40), though this was particularly sensitive to modelling assumptions assessed in our sensitivity analyses, and the multivariable findings suggested longer time to diagnosis for men with heterosexual transmission in particular. For pre‐migration infections, median time from arrival to diagnosis was estimated at 0.60 years (95% CrI 0.47–0.83), and an estimated 50.5% of those diagnosed to date were late.

Conclusion

Late HIV diagnosis represents a significant challenge. Estimating time‐to‐diagnosis distributions may help separate the influence of testing delays from incidence and migration. The burden demonstrated across multiple groups highlights the need for normalization and expansion of HIV testing within and beyond traditional risk groups.

Keywords: delayed diagnosis, epidemiology, HIV, HIV testing, late diagnosis

INTRODUCTION

There is usually no definitive evidence available regarding the timing of infection for new HIV diagnoses. A pragmatic definition of late diagnosis of HIV that is commonly applied to surveillance data requires the presence of an AIDS‐defining event or a CD4 count of <350 cells/μL at time of diagnosis and the absence of any evidence of recent infection [1]. Late HIV diagnoses are a significant clinical and public health concern, given their association with higher morbidity and mortality, increased transmission and greater healthcare system costs [2, 3, 4]. Nearly half of all new diagnoses across 23 European countries were late in a recent study [5], in contrast with the World Health Organization interim target of <20% for the region by 2025 [6]. This gap indicates both a significant challenge and a clear opportunity for action to improve public health.

Understanding the factors contributing to late HIV diagnoses on a national level is important to help inform policy changes and interventions to improve outcomes for individuals and for public healthcare systems. Routine surveillance data are readily available and is commonly used to study factors associated with late diagnosis of HIV. This frequently involves an assessment of the proportion diagnosed late in a given period of surveillance data and related regression analyses. However, the interpretation of these analyses presents several challenges.

A challenge with the commonly used logistic regression model is that the resulting odds ratios may be mistakenly interpreted as relative risks [7, 8]. Alternative approaches, such as modified Poisson or log‐binomial regression, can estimate the relative risk directly. However, these methods come with their own limitations: Modified Poisson models can yield predicted probabilities that are not plausible, while log‐binomial models can have issues with convergence, which can make them difficult to fit in practice [8, 9, 10, 11].

A further challenge relates to the fact that the proportion diagnosed late in a given period is influenced not only by testing delays, but also by changes in incidence and migration. For instance, the proportion diagnosed late may increase over time in one subgroup due to decreasing incidence (with fewer new infections available to be diagnosed), regardless of testing intensity [12]. Similarly, a high proportion of late diagnoses may partly reflect pre‐existing late, but undiagnosed, infections in people arriving into the country, regardless of the speed of testing on arrival. These aspects shape the proportion diagnosed late and make it difficult to separate the influence of testing delays versus changes in incidence or migration, thus complicating the interpretation of surveillance data for the purposes of guiding interventions to reduce late diagnoses.

To address these challenges, we used a flexible Bayesian modelling framework and analysed national HIV surveillance data in Ireland (2015–2024). We demonstrate the application of an approach that combines (i) logistic regression with marginal standardization to estimate prevalence ratios (PRs) of late diagnosis among HIV notifications and (ii) estimation of time‐to‐diagnosis distributions to help distinguish testing delays from differences in incidence and migration. We aimed to improve understanding of the factors associated with late diagnosis of HIV in Ireland to help inform the design of interventions to reduce late diagnoses.

METHODS

Data source and study population

We conducted a retrospective, observational analysis of routinely collected national HIV notifications from 2015 to 2024 using the national computerized infectious disease reporting system for Ireland. HIV is a notifiable disease in Ireland, meaning that medical practitioners (including clinical directors of laboratories) are legally required to notify new diagnoses to the Medical Officer of Health [13]. Eligible cases for the statistical analyses were those aged 15 years or older at the time of notification with a probable mode of transmission of sex between men, heterosexual contact or injecting drug use. We excluded those of younger age or alternative modes of transmission as there were very small numbers in these categories and they may represent different testing routes and pathways as compared with the general adult population with sexual or injecting drug use transmission, which was the focus of this study. Impossible or implausible dates and blood test results were set to missing, as detailed in the Supporting Information. Based on the experience of the study team, the date of notification on the computerized infectious diseases reporting system was more reliable and complete as compared with the recorded diagnosis date available in our dataset, since it is systematically recorded and is used more consistently in routine surveillance reporting. Thus, we used the notification date as the date of diagnosis for the main analyses and assessed the impact of this decision through a sensitivity analysis.

Definitions

A new diagnosis indicator was created to identify individuals newly diagnosed with HIV in Ireland. Individuals were classified as newly diagnosed if they were explicitly recorded as not having previously tested positive for HIV abroad, provided they did not have a previous positive test in Ireland in an earlier year. The statistical analyses were restricted to new diagnoses, as our study was focused on those newly diagnosed in Ireland as opposed to transfers of care for people with existing diagnoses from other countries.

Late diagnosis was defined following the EuroTEST HIV Late Diagnosis Definition Working Group consensus definition [1]. Applying this definition to our dataset, we used a hierarchical coding scheme to classify diagnoses based on available variables:

Those with clinical evidence of AIDS‐defining illness were classified as late.
Those with a CD4 count of less than 350 cells/μL were classified as late, provided they did not have evidence of recent infection. Those with missing data for recent infection were treated as not having recent infection, and the impact of this decision was assessed through a sensitivity analysis as outlined below. Evidence of recent infection was defined as having any of the following recorded:
1. Positive p24 antigen at diagnosis.
2. Previous negative HIV test in the same year as notification or in the year prior to notification.
3. Clinical evidence of acute seroconversion illness.

Statistical analysis

Logistic regression with marginal standardization

We carried out logistic regression using the brms R package [14]. We fit univariable and multivariable models for the five variables of interest: age, area of residence in Ireland, region of origin, combined probable mode of transmission and sex and notification period. Region of origin was based on the region of birth of individuals. This was classified into seven categories: Ireland, Western Europe (excluding Ireland), Eastern Europe, Central Europe, Sub‐Saharan Africa, Latin America and Caribbean and Other. Area of residence in Ireland was categorized based on the six administrative Health Regions [15]. Three of these Health Regions include parts of the capital Dublin and surrounding counties. The other three Health Regions cover parts of the south, west and north‐west of the country, including some areas where a high proportion of people live in rural settings [16]. We combined the variables mode of transmission and sex into a single variable with the following levels: sex between men, heterosexual transmission (men), heterosexual transmission (women) and injecting drug use. This approach was used to improve interpretability of the multivariable model estimates, as some of the transmission categories were sex‐specific; therefore, adjusting for sex and mode of transmission as separate covariates would have resulted in challenges in interpretation of coefficient estimates.

We evaluated several age and time parameterisations, because we were uncertain as to the functional form of their association with late diagnosis from the outset. We therefore assessed whether more flexible approaches or simpler, more interpretable categorisations were more appropriate and compared each option in terms of predictive performance through approximate leave‐one‐out cross‐validation with Pareto smoothed importance sampling (PSIS‐LOO), as implemented in the loo R package [17]. Where model specifications were not significantly different in PSIS‐LOO comparison, we favoured the simpler, more interpretable option. Thus, in the final models, we included age in decades and notification period in categories (2015–2019, 2020–2021, 2022–2024). These periods were chosen to ensure sufficient numbers in each and to allow for changes related to the COVID‐19 pandemic. The 2015–2019 period represents the pre‐COVID‐19 period; 2020–2021 captures the main period of COVID‐19‐related restrictions in Ireland, with potential disruptions to healthcare access and testing; and 2022–2024 represents the subsequent period. Because changes in testing access and other factors may have influenced the probability of late diagnosis in these periods, this categorisation allowed for adjustment for period‐specific differences rather than assuming a linear time relationship. We used weakly informative priors of $Normal (0, 1)$ for the intercept and regression coefficients. The primary estimand was the marginal (population‐averaged) PR for late diagnosis among new diagnoses; this was standardized to the covariate distribution observed among new diagnoses in the study period.

Time‐to‐diagnosis modelling

There were several stages involved in modelling the time to diagnosis. First, we estimated the posterior distribution for each individual's time since infection using the approach developed by Pantazis et al. [18] This model uses parameters from the CASCADE seroconverter cohorts and estimates the time of infection for individuals based on a bivariate linear mixed model for CD4 count and viral load, adjusted for age, sex, region of origin and transmission route. It also incorporates information on AIDS diagnoses through a time to AIDS submodel. While this model allows for biomarker measurements at multiple time points per person, it has also been applied to situations with measurements available at only one time point per person [19], as was the case in our dataset. We adapted the R code and parameters provided in the supplementary material of the original Pantazis et al. publication [18]. The onset of infection risk was defined, as in the Pantazis et al. publication [18], as the maximum of: the date of prior negative HIV test (where available), the date when aged 10 years and the 1st of January 1980. We took the date of the last negative test to be the 1st of January in the recorded year of the previous negative test. For each individual, we found the posterior mode and drew 100 samples using rejection sampling, as implemented by Pantazis et al. [18], creating 100 completed datasets for subsequent analysis.

Next, we classified infections as occurring in Ireland (in‐Ireland infections) or prior to arrival in Ireland (pre‐migration infections). Since year of arrival, but not date of arrival, was available, date of arrival was estimated by sampling from a uniform distribution of arrival dates in each person's year of arrival (with an upper bound at the date of diagnosis, if in the same year as arrival). This was carried out separately for each of the 100 draws of infection dates for each person, to account for the uncertainty in arrival date. Then, for those with region of origin outside Ireland, we classified those with estimated infection date prior to date of arrival in Ireland as pre‐migration infections and those with estimated infection date after arrival date as in‐Ireland infections. We assumed that people with region of origin in Ireland had in‐Ireland infections for the main analysis.

Finally, we modelled the delay distribution separately for pre‐migration and in‐Ireland infections. We defined the start date for the time‐to‐diagnosis modelling as the infection date for in‐Ireland infections, and the arrival date for pre‐migration infections. Thus, for pre‐migration infections, the delay represents time from arrival in Ireland to diagnosis (i.e., post‐arrival interval), not total time from infection to diagnosis. We excluded those with start dates prior to the 1st of January 2015. We applied lognormal regression as implemented in brms, using brm_multiple to fit the model across the 100 completed datasets and combine posterior draws across these fits into a single posterior distribution, from which medians and credible intervals (CrIs) were summarized [14]. To account for right truncation, whereby those not yet diagnosed by the end of the surveillance period are not known, we used the truncation modifier as implemented in brms [14]. The primary estimands were the overall median delay (years), and the ratio of median delays (referred to as ‘median delay ratios’ here) for subgroup contrasts. We included the same covariates as in the logistic regression analysis, but defined the calendar period based on the start date as opposed to the notification date, and age based on age at start date rather than age at notification. We used priors of $Normal (\log (3), 0.75)$ for the intercept, $Normal (0, 1)$ for the regression coefficients and $half - Normal (0, 0.5)$ for the residual standard deviation, all as parameterised in brms [14]. The priors were chosen to reflect a plausible range of values for the median delay based on evidence from previous studies [19, 20, 21].

Missing data handling

Cases with missing data for any of the following variables were excluded from the main statistical analysis: new diagnosis indicator, CD4 count, clinical stage or any of the regression covariates. Cases with missing data for recent infection were treated as not having recent infection for the main analysis. In addition, those with missing year of arrival were excluded from the delay distribution analyses if their region of origin was outside Ireland.

Sensitivity analyses

Missing data

We conducted a sensitivity analysis of the logistic regression findings to the missing data handling decisions by using different multiple imputation approaches. We used fully conditional specification as implemented with the mice R package with two different specifications [22]. For the first, we imputed the new diagnosis indicator, some of the component variables for the late diagnosis definition (clinical stage, CD4 count and recent negative test), and the covariates. For the second, we limited it to known new diagnoses with complete outcome data and only imputed the covariates. In addition, we compared results with substantive model compatible fully conditional specification, as implemented with the smcfcs R package [23].

Diagnosis date

We conducted a sensitivity analysis of the logistic regression findings to the decision to use notification date as the date of diagnosis by repeating the analysis using recorded diagnosis date instead.

Probable country of infection

As a sensitivity analysis for the assumption that those with region of origin in Ireland had in‐Ireland infections, we repeated the time‐to‐diagnosis analysis excluding those with region of origin in Ireland who had clinician‐reported probable country of infection outside Ireland or missing.

Delay analysis priors and distribution

Following review of the primary analysis, we repeated the time‐to‐diagnosis regression analyses using alternative priors and distributions, including exponential and Weibull, to assess the sensitivity of the main findings to some of the modelling assumptions. This was conducted for the absolute estimates of median delay (from intercept‐only fits), and for the subgroup contrasts from multivariable fits.

All data exploration and analyses were conducted using R version 4.5.2 [24]. We report posterior medians and 95% CrIs (equal‐tailed). Model diagnostics included posterior predictive checks and convergence and efficiency diagnostics. All models showed evidence of satisfactory convergence and sampling efficiency as assessed by trace plots, $\hat{R}$ and effective sample size values. Further details on the statistical methods and diagnostics are provided in the Supporting Information.

RESULTS

Descriptive results

The numbers excluded and reasons for exclusion are summarized in Figure 1. Further information on missing data is included in Figure S1. There were 6113 HIV notifications during the study period. Of these, 1952 were in people aged 15 years or older with a new diagnosis and probable mode of transmission of sex between men, heterosexual contact, injecting drug use or missing. CD4 count and clinical stage data were available for a total of 1551 of these people, and 684 (44.1%) of these had late diagnoses. These 1551 cases, including 64 with missing data for some of the regression model covariates, are described by sociodemographic and epidemiologic subgroups and late diagnosis status in Table 1. In some groups, greater than 50% of the diagnoses in the study period were late; this included people aged over 50 years, people whose probable mode of transmission was heterosexual contact, people born in Eastern Europe or Sub‐Saharan Africa, and people residing in the South West or West and North West areas of Ireland.

Flowchart of case exclusion for the statistical analyses. Note that further exclusions were applied prior to the different delay distribution regression fits, as outlined in Methods and Results.

TABLE 1.

New HIV diagnoses in Ireland by late diagnosis status, 2015–2024.

Variable	Level	Non‐late diagnosis		Late diagnosis
Variable	Level	n = 867		n = 684
Age group	15–29	294	70.0%	126	30.0%
	30–50	494	52.9%	439	47.1%
	>50	79	39.9%	119	60.1%
Sex	Male	702	57.5%	518	42.5%
Sex	Female	165	49.8%	166	50.2%
Probable mode of transmission	Sex between men	558	64.5%	307	35.5%
	Heterosexual	263	45.4%	316	54.6%
	Injecting drug use	36	51.4%	34	48.6%
	Missing	10	27.0%	27	73.0%
Region of origin	Ireland	332	57.2%	248	42.8%
	Western Europe (excluding Ireland)	34	52.3%	31	47.7%
	Eastern Europe	32	45.1%	39	54.9%
	Central Europe	56	63.6%	32	36.4%
	Sub‐Saharan Africa	167	47.6%	184	52.4%
	Latin America and Caribbean	193	62.9%	114	37.1%
	Other	35	60.3%	23	39.7%
	Missing	18	58.1%	13	41.9%
Area of residence	Dublin and North East	252	55.3%	204	44.7%
	Dublin and Midlands	328	60.6%	213	39.4%
	Dublin and South East	121	59.3%	83	40.7%
	South West	57	47.5%	63	52.5%
	Mid‐West	48	52.2%	44	47.8%
	West and North West	61	44.2%	77	55.8%
Notification period	2015–2019	496	57.1%	372	42.9%
	2020–2021	96	51.9%	89	48.1%
	2022–2024	275	55.2%	223	44.8%

Open in a new tab

The median CD4 count in these 1551 individuals was 346 cells/μL (inter‐quartile range 176–527, range 0–2974). A total of 185 individuals had clinical evidence of an AIDS‐defining illness. Of 784 individuals with a CD4 count of less than 350 cells/μL, 112 were classified as non‐late diagnoses due to having one or more indicators of recent infection (positive p24 antigen [n = 19], clinical evidence of acute seroconversion illness [n = 21] and/or recent prior negative test [n = 99]). The trends in numbers and percentages of late diagnoses over the study period are summarized in Figure 2.

New HIV diagnoses in Ireland, 2015–2024.

Logistic regression

Of the 1551 new diagnoses described in Table 1, 64 had missing data for one or more input variables and so were excluded from the logistic regression, leaving 1487 cases for the regression fits. Table 2 presents marginal (population‐averaged) PRs for late diagnosis among new diagnoses. The multivariable model included all five input variables: age, probable mode of transmission and sex, region of origin, area of residence and notification period. Increasing age was associated with higher PR, which was similar in both univariable and multivariable models (multivariable PR: 1.20; 95% CrI 1.15–1.26; per decade increase in age). The region of origin contrasts used Ireland as the reference level. Sub‐Saharan African origin was associated with a higher PR in the univariable model (PR: 1.25; 95% CrI 1.09–1.43); this association was attenuated in the multivariable model but still in the same direction (PR: 1.14; 95% CrI 0.95–1.36). In contrast, Latin American and Caribbean origin was associated with greater PR in the multivariable model only (univariable PR: 0.88; 95% CrI 0.73–1.05; multivariable PR: 1.21; 95% CrI 1.03–1.42). In terms of probable mode of transmission and sex, heterosexual transmission was associated with greater PR as compared with sex between men, particularly for heterosexual men (univariable PR: 1.62; 95% CrI 1.42–1.86. Multivariable PR: 1.43; 95% CrI 1.23–1.67). Injecting drug use was also associated with a higher PR compared with sex between men, but with a wider CrI (multivariable PR 1.25; 95% CrI 0.94–1.58). The area of residence comparisons took Dublin and North East as the reference group. We found that West and North West had a higher PR in both models, though this was attenuated in the multivariable fit (univariable PR: 1.24; 95% CrI 1.03–1.48; multivariable PR: 1.18; 95% CrI 0.97–1.42). Similar PRs were reported for the different notification periods. Posterior predictive checks were satisfactory, as shown in Figures S2 and S3.

TABLE 2.

Univariable and multivariable marginal (population‐averaged) prevalence ratios for late diagnosis among new diagnoses, estimated from logistic regression models with marginal standardization over the observed covariate distribution.

Variable	Level	Univariable	Multivariable
Age (per 10 years)	+10 years	1.22 (1.17, 1.27)	1.20 (1.15, 1.26)
Probable mode of transmission and sex	Sex between men	Ref	Ref
	Heterosexual transmission (men)	1.62 (1.42, 1.86)	1.43 (1.23, 1.67)
	Heterosexual transmission (women)	1.40 (1.21, 1.61)	1.28 (1.06, 1.53)
	Injecting drug use	1.28 (0.97, 1.63)	1.25 (0.94, 1.58)
Region of origin	Ireland	Ref	Ref
	Western Europe (excluding Ireland)	1.11 (0.81, 1.42)	1.22 (0.92, 1.55)
	Eastern Europe	1.25 (0.96, 1.55)	1.11 (0.82, 1.43)
	Central Europe	0.83 (0.61, 1.10)	0.98 (0.74, 1.27)
	Sub‐Saharan Africa	1.25 (1.09, 1.43)	1.14 (0.95, 1.36)
	Latin America and Caribbean	0.88 (0.73, 1.05)	1.21 (1.03, 1.42)
	Other	0.92 (0.65, 1.24)	1.07 (0.77, 1.41)
Area of residence	Dublin and North East	Ref	Ref
	Dublin and Midlands	0.90 (0.78, 1.04)	0.96 (0.83, 1.11)
	Dublin and South East	0.89 (0.72, 1.08)	0.92 (0.75, 1.11)
	South West	1.13 (0.92, 1.37)	1.11 (0.89, 1.36)
	Mid‐West	1.05 (0.82, 1.33)	1.06 (0.82, 1.32)
	West and North West	1.24 (1.03, 1.48)	1.18 (0.97, 1.42)
Notification period	2015–2019	Ref	Ref
	2020–2021	1.11 (0.93, 1.32)	1.07 (0.90, 1.26)
	2022–2024	1.07 (0.94, 1.21)	0.95 (0.84, 1.08)

Open in a new tab

Note: The multivariable model included the following variables: age, probable mode of transmission and sex, region of origin, area of residence and notification period. Prevalence ratio (95% credible interval [CrI]); n = 1487.

Time‐to‐diagnosis distributions

For the time‐to‐diagnosis analysis, those with region of origin outside Ireland and missing year of arrival were also excluded (n = 121), leaving 1366 available for classification. Across the 100 imputed infection time datasets, a median of 913 were classified as having in‐Ireland infections and 453 as having pre‐migration infections. We then excluded a median of 336 in‐Ireland infections with infection dates prior to 2015 and a median of 50 pre‐migration infections with arrival dates prior to 2015. This left a median of 577 in‐Ireland infections and 403 pre‐migration infections for the time‐to‐diagnosis distribution modelling. A summary of the overall estimates from the time‐to‐diagnosis distribution modelling is shown in Table 3. Of those diagnosed to date, a median of 25.7% of in‐Ireland infections were classified as late, as compared with 50.5% of pre‐migration infections. We refer to our estimates of the underlying untruncated distribution, after accounting for the truncated nature of the observed data, as ‘truncation‐adjusted’ here. The truncation‐adjusted median time from infection to diagnosis for in‐Ireland infections was estimated at 1.94 years (95% CrI 1.36–3.40 years) from an intercept‐only fit, though this was particularly sensitive to the assumed distribution of delays, as described in further detail in the sensitivity analysis findings below. The truncation‐adjusted median time from arrival to diagnosis for pre‐migration infections was estimated at 0.60 years (95% CrI 0.47–0.83 years) from an intercept‐only fit; time from infection to diagnosis was not estimated for this group.

TABLE 3.

Estimated median and 95% credible interval (CrI) of time to diagnosis, from intercept‐only lognormal models.

Infection location	n ^a	% Late ^a	Delay measured from	Median delay, years (95% CrI)
Infection location	n ^a	% Late ^a	Delay measured from	Unadjusted	Truncation‐adjusted
In Ireland	577	25.7%	Infection to diagnosis	0.80 (0.69, 0.92)	1.94 (1.36, 3.40)
Pre‐migration	403	50.5%	Arrival to diagnosis	0.38 (0.32, 0.44)	0.60 (0.47, 0.83)

Open in a new tab

Note: Shown separately for in‐Ireland infections (time from infection to diagnosis) and pre‐migration infections (time from arrival to diagnosis).

^{^a}

Median values across 100 imputations.

Table 4 shows median delay ratios from the lognormal regression models with right‐truncation adjustment. The multivariable models again included all five input variables: age, probable mode of transmission and sex, region of origin, area of residence and infection or arrival period. For in‐Ireland infections, increasing age was associated with longer time to diagnosis in the univariable model (median delay ratio per decade 1.38; 95% CrI 1.10–1.83), with some attenuation in the multivariable model (1.19; 95% CrI 0.94–1.56). Regarding probable mode of transmission, as compared with sex between men, heterosexual transmission among men in particular was associated with longer delays in the multivariable analysis (multivariable median delay ratio 3.39; 95% CrI 1.48–8.96). For the pre‐migration infections, the CrIs all included 1 after multivariable adjustment. Plots of the delay distributions and posterior predictive checks are available in Figures S4–S7.

TABLE 4.

Posterior estimates from the lognormal regression model, fitted separately for in‐Ireland and pre‐migration infections.

Variable	Level	Infection to diagnosis		Arrival to diagnosis
Variable	Level	Univariable	Multivariable	Univariable	Multivariable
Age (per 10 years)	+10 years	1.38 (1.10, 1.83)	1.19 (0.94, 1.56)	0.90 (0.70, 1.16)	0.84 (0.62, 1.13)
Probable mode of transmission and sex	Sex between men	Ref	Ref	Ref	Ref
	Heterosexual transmission (men)	4.15 (1.88, 10.71)	3.39 (1.48, 8.96)	1.20 (0.65, 2.22)	2.08 (0.95, 4.77)
	Heterosexual transmission (women)	2.25 (1.02, 5.22)	2.06 (0.83, 5.27)	1.04 (0.63, 1.77)	1.83 (0.85, 4.10)
	Injecting drug use	0.78 (0.28, 2.17)	0.66 (0.23, 1.91)	0.67 (0.20, 2.38)	0.69 (0.18, 2.72)
Region of origin	Ireland	Ref	Ref
	Western Europe (excluding Ireland)	0.60 (0.20, 1.92)	0.71 (0.24, 2.23)	Ref	Ref
	Eastern Europe	2.29 (0.60, 9.17)	1.92 (0.49, 7.98)	2.11 (0.72, 6.63)	1.83 (0.56, 6.14)
	Central Europe	0.75 (0.28, 2.21)	0.84 (0.32, 2.39)	1.23 (0.40, 3.87)	1.20 (0.38, 3.83)
	Sub‐Saharan Africa	1.12 (0.51, 2.49)	0.68 (0.27, 1.68)	0.69 (0.33, 1.48)	0.58 (0.24, 1.39)
	Latin America and Caribbean	0.43 (0.23, 0.77)	0.57 (0.31, 1.04)	0.91 (0.41, 2.04)	1.10 (0.46, 2.60)
	Other	0.48 (0.17, 1.50)	0.58 (0.21, 1.78)	1.40 (0.46, 4.36)	1.52 (0.49, 4.90)
Area of residence	Dublin and North East	Ref	Ref	Ref	Ref
	Dublin and Midlands	0.79 (0.40, 1.45)	0.88 (0.46, 1.59)	1.69 (1.00, 2.82)	1.59 (0.91, 2.78)
	Dublin and South East	0.83 (0.39, 1.70)	0.94 (0.45, 1.90)	1.34 (0.63, 2.96)	1.31 (0.59, 2.98)
	South West	2.30 (0.81, 7.25)	1.89 (0.67, 5.88)	2.54 (0.95, 6.99)	2.15 (0.75, 6.04)
	Mid‐West	1.04 (0.37, 3.03)	1.03 (0.37, 2.98)	1.88 (0.73, 4.79)	1.74 (0.64, 4.58)
	West and North West	2.15 (0.73, 7.00)	1.85 (0.66, 5.89)	2.87 (1.18, 7.58)	2.29 (0.89, 6.34)
Start period	2015–2019	Ref	Ref	Ref	Ref
	2020–2021	1.88 (0.88, 4.36)	1.81 (0.89, 3.99)	1.14 (0.54, 2.47)	1.07 (0.50, 2.37)
	2022–2024	1.10 (0.52, 2.64)	0.97 (0.47, 2.21)	1.41 (0.83, 2.69)	1.32 (0.77, 2.53)

Open in a new tab

Note: Reported as median delay ratios from infection or arrival to diagnosis. The multivariable models again included all five input variables: age, probable mode of transmission and sex, region of origin, area of residence and infection or arrival period. Ratios of median delay (95% credible interval [CrI]) from lognormal regression models with right‐truncation adjustment. Median n across imputations: in‐Ireland = 577; pre‐migration = 403.

Sensitivity analyses

PRs estimated from the logistic regression with marginal standardization after multiple imputation are shown in Table S1. The findings for age and heterosexual transmission remain similar with any of the approaches tested, with small changes in the magnitude of effect and CrIs seen for some contrasts. The comparison of PRs using recorded diagnosis date as compared with notification date showed similar results, as shown in Table S2. Comparisons of observed and imputed values are shown in Figures S8 and S9. Regarding the delay distribution probable country of infection sensitivity analysis, again the substantive findings were similar, as shown in Table S3.

In terms of the sensitivity analyses for the delay distribution priors and model, the estimates for median time to diagnosis were sensitive to modelling assumptions. The posterior estimates for the residual standard deviation with the primary analysis were 1.77 (95% CrI 1.53–2.10) for in‐Ireland infections and 1.63 (95% CrI 1.44–1.87) for pre‐migration infections. With different lognormal prior specifications, including broader priors for the residual standard deviation, the median delay estimates varied from 1.94 to 2.16 for in‐Ireland infections and from 0.60 to 0.63 for pre‐migration infections (Table S4). The median delay estimate for in‐Ireland infections was more sensitive to distributional family, with the posterior estimates for the exponential and Weibull distributions being 1.38 and 1.37, respectively. In terms of the subgroup comparisons, the finding of a longer delay ratio for heterosexual transmission among men with in‐Ireland infections was consistent across the different modelling specifications. The other delay ratios were generally in the same direction and of similar magnitude, with some differences particularly in the pre‐migration analysis, as shown in further detail in Tables S5 and S6.

DISCUSSION

Key findings and comparison with other evidence

Late HIV diagnosis remains a challenge in Ireland, as in many other European countries and contributes to potentially avoidable morbidity, mortality and onward transmission risk. Overall, 44.1% of included new diagnoses with available information were classified as late, slightly below the 49.6% described in a recent multi‐country European study and slightly above the 39.4% reported in a recent study in England [5, 25]. This remains far above the WHO interim target of less than 20% for the European region by 2025 [6]. Factors associated with late diagnosis included increasing age and heterosexual transmission. The delay distribution modelling should be interpreted in two distinct contexts. First, for infections likely acquired in Ireland, in univariable models, we found evidence of an association between increasing age and heterosexual transmission with longer delays, and in multivariable models, we found evidence of an association between heterosexual transmission in men with longer delays in particular. This suggests that the association with late diagnosis seen for some subgroups may at least partly be related to testing delays within Ireland and not only to differences in incidence or migration. This highlights opportunities for improvement in national testing strategies to address potential gaps or barriers for certain population groups. Second, for likely pre‐migration infections, late diagnosis may reflect infections that were already at a relatively advanced stage before arrival in Ireland. Although many in this group were diagnosed within 1 year of arrival, they still frequently met the late diagnosis definition, highlighting the need for prompt testing on arrival to adequately reach migrant populations and ensure rapid diagnosis and immediate access to treatment. The recent Irish National Sexual Health Strategy and upcoming HIV Action Plan provide opportunities to prioritize evidence‐informed actions to tackle late HIV diagnosis [26].

Several of the key findings are consistent with evidence from other European countries. Heterosexual transmission has frequently been reported to be associated with late diagnosis in Europe [5, 25, 27, 28, 29]. Lower proportions of late diagnoses in GBMSM in Europe may be related to more frequent testing and greater risk awareness [30]. Although heterosexual transmission was associated with a higher relative risk of late diagnosis and longer testing delays in our study, the absolute numbers of late diagnoses were similar for sex between men and heterosexual transmission, highlighting the over‐representation of GBMSM in late and overall HIV diagnoses [31]. European standards recommend a range of HIV testing approaches, including routine testing for key population groups, community‐based testing, self‐testing options, routine testing for those attending high HIV prevalence healthcare settings, and testing for people with indicator conditions (conditions associated with higher prevalence of undiagnosed HIV, such as certain infections, cancers, skin conditions and blood abnormalities) [32]. Our findings highlight the importance of such diverse testing approaches by demonstrating that HIV is not limited to one population group, supporting the need for the normalization and expansion of HIV testing across various settings. Indicator condition testing is an example of an intervention that may reach those outside current focused testing in STI clinics or other targeted interventions. In addition to increasing early diagnoses, indicator condition testing can also reduce further diagnostic delay by identifying later‐stage infections that had previously been missed. Limited local evidence highlights some of the challenges associated with indicator condition testing in Ireland [33, 34], suggesting potential opportunities to strengthen and expand this testing approach.

Increasing age is another factor that has been identified across several other European countries [5, 25, 27, 29, 35, 36, 37, 38]. Potential barriers for testing in older adults may include low perceived need by patients and clinicians [39]. We found that increasing age was associated with late diagnosis in the logistic regression and with longer delays in the univariable delay distribution modelling, though the delay association was attenuated after multivariable adjustment. This suggests a potential opportunity to provide earlier and more widespread testing for older adults in Ireland. However, it is important to also highlight that the majority of late diagnoses were in the 30–50‐year age group, reflecting the larger number of total diagnoses in that age bracket and ongoing need for testing and interventions across all ages.

Although most of the late diagnoses, in absolute terms, occurred in regions near Dublin, we found evidence of a higher proportion of late diagnoses among those residing in the West and North West or in the South West of Ireland. The multivariable estimates were attenuated, suggesting that different demographic profiles may partly account for this, and the multivariable delay distribution modelling did not provide conclusive evidence of differences in delays by area of residence. Differences in HIV‐related knowledge, attitudes and testing in other settings have been associated with factors such as rural geographic locations and socioeconomic deprivation [40, 41]. Focused regional analyses could build on our findings to understand the contextual factors underlying the observed regional differences in Ireland.

Multiple other studies have demonstrated a higher proportion of late diagnoses among migrants in Europe, especially among people born in Africa [27, 29, 36, 38, 42]. We similarly found a higher PR for late diagnosis among people born in sub‐Saharan Africa, particularly in the univariable estimates. The change in PR among people born in Latin America and the Caribbean between univariable and multivariable models may partly reflect adjustment for a younger age and higher proportion GBMSM in this group, and the elevated multivariable PR may also partly reflect the impact of pre‐migration infections. Barriers to HIV testing among migrants in Europe may include stigma, discrimination, language barriers and cultural barriers [43]. Although the median time from arrival to diagnosis was relatively short for those with likely pre‐migration infections in our analysis, the high proportion diagnosed late in this group highlights the global nature of this challenge and the need for a coordinated international response alongside domestic interventions in order to meet WHO targets. It also demonstrates the importance of early testing and rapid linkage to care for migrants in Ireland, following recommendations in European and Irish guidance for non‐stigmatizing and culturally sensitive interventions for this key priority group [44, 45]. Finally, it is important to recognize that this priority group comprises multiple distinct subgroups with different circumstances and needs, depending on factors such as cultural and demographic background, language and reason for migration. Each subgroup may therefore experience different barriers and challenges, requiring specific and tailored interventions.

Strengths and limitations

In this study, we extend standard late HIV diagnosis analyses by combining Bayesian logistic regression with marginal standardization and truncation‐adjusted delay distribution modelling. Bayesian logistic regression with marginal standardization provides a flexible alternative to modified Poisson regression for estimating PRs when the outcome is common [46, 47], with the advantages of constraining the predicted probabilities to plausible values, making the target population explicit [11], and allowing for transparent propagation of uncertainty and extension to more complex scenarios in a Bayesian framework [47]. Regarding the time‐to‐diagnosis estimates, the model used here has been shown to have satisfactory performance for pre‐ versus post‐migration classification even when most individuals only contribute a single CD4 measurement, which is a common constraint in national surveillance system data [18, 19]. We build on this line of work by using the estimated infection times for a lognormal regression of time to diagnosis that explicitly accounts for the truncated nature of the surveillance data. This allows for the estimation of covariate‐specific summaries of the underlying delay distributions, such as median delay ratios, while taking advantage of the Bayesian framework to propagate uncertainty from preceding stages through to these final estimates.

The delay distribution modelling applied here also has several limitations and assumptions. Since estimates were obtained by separately refitting the regression over 100 sampled datasets, the resulting inferences should be viewed as approximate. The model assumes that the parameters estimated from the CASCADE seroconverter cohorts can be used to approximate biomarker and AIDS progression in our population. The modelling approach also assumes that, conditional on the included covariates, there are no additional unmodelled time‐varying changes in testing delay. Some of the findings were sensitive to the model specifications assessed in our sensitivity analysis. The median delay for in‐Ireland infections was particularly sensitive to the choice of family, with exponential and Weibull models providing lower estimates (but overlapping CrIs) compared to the lognormal. The sensitivity of delay estimation to choice of parametric family is a known issue in studying epidemiologic delays [48]. This is not unexpected in our study, particularly as our observed data are truncated. The finding of a longer delay ratio for heterosexual men with in‐Ireland infections was consistent across the different model specifications. Other subgroup delay ratios were generally in the same direction across the comparisons, with some differences in the estimates as noted in the Supporting Information. However, all of these specifications still include several simplifying assumptions. For example, they assumed additive covariate effects, did not allow for interactions and assumed any additional shape or dispersion parameters were constant across subgroups. Posterior predictive checks showed the fit to the observed data appeared reasonable, but ultimately the underlying untruncated delay distribution cannot be directly verified from the available data, and there may be subgroups with prolonged delays that are not well modelled and identified by this approach.

Our analysis is limited to routinely available variables in surveillance data. These capture important demographic and clinical markers, but can only indirectly reflect the individual and service‐level barriers to testing, which may include fear, stigma, low risk perception and difficulties accessing healthcare [49]. Qualitative research and linkage of surveillance data to testing, behavioural and clinical information may help identify contextually relevant barriers for particular groups with longer delays in Ireland. More detailed laboratory data, such as using viral sequence genetic diversity, may provide more accurate estimates of infection times as compared with methods primarily based on CD4 measurements [50, 51]. We focused on the latter at this time since this reflects the routinely available data in our surveillance system, but additional information could be incorporated in the future.

The proportion of HIV notifications included in the statistical analyses should be interpreted in the context of the study target population. The primary analyses focused on known new diagnoses, because the aim was to examine late diagnosis and time to diagnosis among people newly diagnosed in Ireland. People with a previous HIV diagnosis, such as those diagnosed prior to arrival in Ireland, represent an important but distinct population. The public health questions for people with pre‐existing diagnoses may relate more to issues of transfer of care, linkage to services and treatment continuity. This group could be examined separately in future work. In addition, for the time‐to‐diagnosis modelling, we restricted our analysis to infections (or arrivals) estimated to have occurred from 2015 onwards. This was done to explicitly define a clearly interpretable time period for assessing testing system delays in Ireland. However, this resulted in a smaller sample size which may contribute to greater uncertainty in the time‐to‐diagnosis estimates.

Finally, missing data are a common challenge in HIV surveillance data in Europe [52]. In this study, missingness affected not only the regression input variables, but also variables used to define eligibility for the main analyses and to construct the late diagnosis and delay outcomes. In particular, cases with missing new diagnosis status, CD4 count or clinical stage were excluded from the logistic regression, with a further 64 individuals excluded due to missing covariates. The multiple imputation sensitivity analyses partly addressed these issues for the logistic regression models, including a full multiple imputation approach that imputed new diagnosis status, components of the late diagnosis definitionand the covariates. The logistic regression findings were broadly similar across the imputation approaches, but these comparisons rely on the missing at random assumption which may not always be applicable. For the delay distribution analysis, individuals with region of origin outside Ireland and missing year of arrival were also excluded, because year of arrival was needed to distinguish between likely pre‐migration infections and in‐Ireland infections. The delay distribution estimates may therefore be more sensitive to missingness in migration‐related variables. The diagnosis‐based surveillance data utilized here also do not provide information on individuals who emigrate or die before diagnosis. Further work could extend the Bayesian framework to jointly model missing covariates and infection or arrival time uncertainty. Continued efforts to improve surveillance data completeness for the key variables studied here would also strengthen future analyses. Additional associations may also become apparent with larger datasets.

CONCLUSION

Using national HIV surveillance data for Ireland (2015–2024), we found that 44.1% of included new diagnoses with available information were late. Among new diagnoses, increasing age and heterosexual transmission were associated with late diagnosis in both univariable and multivariable analyses. By combining Bayesian logistic regression with marginal standardization and truncation‐adjusted modelling of time to diagnosis, we demonstrate an extension of routine surveillance analyses in an attempt to separate testing delays from changes in incidence and migration. Our estimates suggest that longer delays from infection in Ireland to diagnosis contribute to the higher proportion diagnosed late in some subgroups, with the strongest evidence in the multivariable estimates seen for men with heterosexual transmission. Individuals with likely pre‐migration infections were often diagnosed within 1 year of arrival but still frequently met the late diagnosis definition. Incorporating delay distribution modelling alongside standard surveillance outputs may help inform targeted testing and other public health interventions to reduce late diagnoses. The burden of late diagnosis across multiple population groups shown here demonstrates the need for further normalization and expansion of HIV testing both within and beyond traditional risk groups.

AUTHOR CONTRIBUTIONS

All authors contributed to study design, interpretation of findings, and review and editing of the manuscript. CDC conducted the data analysis and wrote the initial draft manuscript. KOB and CW provided additional statistical input and review. DI supervised the study.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

ETHICS STATEMENT

This analysis was conducted as part of public health work using routine surveillance data and did not require research ethics committee approval.

Supporting information

Figure S1. Missingness pattern among new HIV diagnoses, Ireland 2015–2024.

Figure S2. Posterior predictive check for the primary logistic regression model showing observed and predicted late diagnosis proportion by covariate groups.

Figure S3. Posterior predictive check for the primary logistic regression model showing binned error plot.

Figure S4. Distribution of estimated delays from infection to diagnosis for infections estimated to have occurred in Ireland with infection dates from 2015 to 2024. Results from the first imputed dataset shown.

Figure S5. Distribution of estimated delays from arrival in Ireland to diagnosis for infections estimated to have occurred prior to arrival in Ireland, for people with arrival dates from 2015 to 2024. Results from the first imputed dataset shown.

Figure S6. Posterior predictive check for the time‐to‐diagnosis model for infections occurring in Ireland (infection to diagnosis). The dark blue line shows the density of estimated delays in the first imputed dataset. Each light blue line shows the density of delays simulated from the fitted multivariable model based on observed predictor values from the first imputed dataset (100 simulations shown).

Figure S7. Posterior predictive check for the time‐to‐diagnosis model for infections occurring prior to arrival in Ireland (arrival to diagnosis). The dark blue line shows the density of estimated delays in the first imputed dataset. Each light blue line shows the density of delays simulated from the fitted multivariable model based on observed predictor values from the first imputed dataset (100 simulations shown).

Table S1. Prevalence ratios (95% CrI) for late diagnosis among new diagnoses, estimated from multivariable logistic regression models with marginal standardization over the observed covariate distribution across different multiple imputation approaches.

Table S2. Prevalence ratios (95% CrI) for late diagnosis among new diagnoses, estimated from multivariable logistic regression models with marginal standardization, using notification date versus diagnosis date.

Table S3. Median delay ratios (95% CrI) from multivariable time‐to‐diagnosis model for in‐Ireland infections, comparing primary analysis with sensitivity analysis where those with region of origin in Ireland who had clinician‐reported probable country of infection outside Ireland or missing were excluded.

Table S4. Sensitivity analysis of estimated median and 95% CrI of time to diagnosis, from intercept‐only models. Shown separately for in‐Ireland infections (time from infection to diagnosis) and pre‐migration infections (time from arrival to diagnosis).

Table S5. Sensitivity analysis of posterior estimates from the multivariable regression models for in‐Ireland infections. Reported as median delay ratios from infection to diagnosis. The multivariable models included all five input variables: age, probable mode of transmission and sex, region of origin, area of residence and infection period.

Table S6. Sensitivity analysis of posterior estimates from the multivariable regression models for pre‐migration infections. Reported as median delay ratios from arrival to diagnosis. The multivariable models included all five input variables: age, probable mode of transmission and sex, region of origin, area of residence and arrival period.

Figure S8. Comparison of imputed versus observed CD4 count distributions by new diagnosis status. Results from 10 imputed datasets shown.

Figure S9. Comparison of imputed versus observed distributions for categorical variables, among those with new diagnoses. Results from 10 imputed datasets shown.

HIV-27-1881-s001.docx^{(736.6KB, docx)}

ACKNOWLEDGEMENTS

The authors thank Nikos Pantazis for providing helpful clarifications regarding the previously published delay estimation model. The authors thank ECDC colleagues Anastasia Pharris, Disa Hansson, Juliana Reyes‐Urueña and Teymur Noori for their helpful insights in the early stages of this project. Claude (Anthropic) prompts were used when drafting and refining parts of the R code used in the analysis, with no access to the data.

DATA AVAILABILITY STATEMENT

The R code used for the analyses is available on Zenodo (https://doi.org/10.5281/zenodo.22177893). The individual‐level data used in this study contain confidential health information and cannot be shared publicly.

REFERENCES

1. Croxford S, Stengaard AR, Brännström J, et al. Late diagnosis of HIV: an updated consensus definition. HIV Med. 2022;23(11):1202‐1208. [DOI] [PMC free article] [PubMed] [Google Scholar]
2. Guaraldi G, Zona S, Menozzi M, et al. Late presentation increases risk and costs of non‐infectious comorbidities in people with HIV: an Italian cost impact study. AIDS Res Ther. 2017;14(1):8. [DOI] [PMC free article] [PubMed] [Google Scholar]
3. Late Presentation Working Groups in EuroSIDA and COHERE . Estimating the burden of HIV late presentation and its attributable morbidity and mortality across Europe 2010–2016. BMC Infect Dis. 2020;20(1):728. [DOI] [PMC free article] [PubMed] [Google Scholar]
4. Mocroft A, Lundgren JD, Sabin ML, et al. Risk factors and outcomes for late presentation for HIV‐positive persons in Europe: results from the Collaboration of Observational HIV Epidemiological Research Europe Study (COHERE). PLoS Med. 2013;10(9):e1001510. [DOI] [PMC free article] [PubMed] [Google Scholar]
5. Reyes‐Urueña J, Stoppa G, Pizzolato F, Marrone G, Hansson D, EU/EEA HIV Network . Late HIV diagnosis: trends, risk factors, and progress toward the 2025 target of <20% late diagnosis in 23 EU/EEA countries, 2022 to 2024. Euro Surveill. 2025;30(47):2500855. [DOI] [PMC free article] [PubMed] [Google Scholar]
6. World Health Organization Regional Office for Europe (WHO/Europe) . Regional Action Plans for Ending AIDS and the Epidemics of Viral Hepatitis and Sexually Transmitted Infections, 2022–2030. WHO/Europe; 2023. [Google Scholar]
7. Greenland S. Noncollapsibility, confounding, and sparse‐data bias. Part 1: the oddities of odds. J Clin Epidemiol. 2021;138:178‐181. [DOI] [PubMed] [Google Scholar]
8. McNutt L‐A. Estimating the relative risk in cohort studies and clinical trials of common outcomes. Am J Epidemiol. 2003;157(10):940‐943. [DOI] [PubMed] [Google Scholar]
9. Greenland S, Maldonado G. The interpretation of multiplicative‐model parameters as standardized parameters. Stat Med. 1994;13(10):989‐999. [DOI] [PubMed] [Google Scholar]
10. Zou G. A modified Poisson regression approach to prospective studies with binary data. Am J Epidemiol. 2004;159(7):702‐706. [DOI] [PubMed] [Google Scholar]
11. Muller CJ, MacLehose RF. Estimating predicted probabilities from logistic regression: different methods correspond to different target populations. Int J Epidemiol. 2014;43(3):962‐970. [DOI] [PMC free article] [PubMed] [Google Scholar]
12. Xia Q, Kobrak P, Wiewel EW, Torian LV. The high proportion of late HIV diagnoses in the USA is likely to stay: findings from a mathematical model. AIDS Care. 2015;27(2):206‐212. [DOI] [PubMed] [Google Scholar]
13. HSE – Health Protection Surveillance Centre . Notifiable Diseases. HPSC; 2025. https://www.hpsc.ie/p/notifiablediseases/listofnotifiablediseases. Accessed February 2, 2026. [Google Scholar]
14. Bürkner PC. Bayesian item response modeling in R with brms and Stan. J Stat Softw. 2021;100(5):1‐54. [Google Scholar]
15. Department of Health . HSE Health Regions. Government of Ireland; 2025. https://www.gov.ie/en/department-of-health/publications/hse-health-regions/. Accessed February 2, 2026. [Google Scholar]
16. McHenry H, Frost D. WDC Census 2022 Summary Report – Highlights for Western Region. Western Development Commission Policy Analysis Team; 2023. [Google Scholar]
17. Vehtari A, Gabry J, Magnusson M, et al. loo: efficient leave‐one‐out cross‐validation and WAIC for bayesian models. 2024. https://mc-stan.org/loo/. Accessed February 2, 2026.
18. Pantazis N, Thomadakis C, del Amo J, et al. Determining the likely place of HIV acquisition for migrants in Europe combining subject‐specific information and biomarkers data. Stat Methods Med Res. 2019;28(7):1979‐1997. [DOI] [PubMed] [Google Scholar]
19. Pantazis N, Rosinska M, van Sighem A, et al. Discriminating between premigration and postmigration HIV acquisition using surveillance data. J Acquir Immune Defic Syndr. 2021;88(2):117‐124. [DOI] [PubMed] [Google Scholar]
20. Van Sighem A, Pharris A, Quinten C, Noori T, Amato‐Gauci AJ, ECDC HIV/AIDS Surveillance and Dublin Declaration Monitoring Networks . Reduction in undiagnosed HIV infection in the European Union/European Economic Area, 2012 to 2016. Euro Surveill. 2017;22(48):17‐00771. [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Gbadamosi SO, Trepka MJ, Dawit R, Jebai R, Sheehan DM. A systematic review and meta‐analysis to estimate the time from HIV infection to diagnosis for people with HIV. AIDS Rev. 2022;24(1):32‐40. [DOI] [PMC free article] [PubMed] [Google Scholar]
22. van Buuren S, Groothuis‐Oudshoorn K. Mice: multivariate imputation by chained equations in r. J Stat Softw. 2011;45(3):1‐67. [Google Scholar]
23. Bartlett J, Keogh R, Bonneville EF, van der Burg L. smcfcs: Multiple imputation of covariates by substantive model compatible fully conditional specification. 2025, https://CRAN.R-project.org/package=smcfcs. Accessed February 2, 2026.
24. R Core Team . R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2025. [Google Scholar]
25. Jasperse J, Harris RJ, Humphreys C, et al. Risk factors for late HIV diagnosis in England, 2015–2023. HIV Med. 2025;27(3):409‐419. [DOI] [PMC free article] [PubMed] [Google Scholar]
26. Department of Health . The National Sexual Health Strategy 2025–2035. Government of Ireland; 2025. [Google Scholar]
27. Bath RE, Emmett L, Verlander NQ, Reacher M. Risk factors for late HIV diagnosis in the East of England: evidence from national surveillance data and policy implications. Int J STD AIDS. 2019;30(1):37‐44. [DOI] [PubMed] [Google Scholar]
28. Noble G, Okpo E, Tonna I, Fielding S. Factors associated with late HIV diagnosis in North‐East Scotland: a six‐year retrospective study. Public Health. 2016;139:36‐43. [DOI] [PubMed] [Google Scholar]
29. Op De Coul ELM, Van Sighem A, Brinkman K, et al. Factors associated with presenting late or with advanced HIV disease in the Netherlands, 1996–2014: results from a national observational cohort. BMJ Open. 2016;6(1):e009688. [DOI] [PMC free article] [PubMed] [Google Scholar]
30. Parczewski M, Gökengin D, Sullivan A, et al. Control of HIV across the WHO European region: progress and remaining challenges. Lancet Reg Health Eur. 2025;52:101243. [DOI] [PMC free article] [PubMed] [Google Scholar]
31. HSE – Health Protection Surveillance Centre . HIV Slideset 2024. HPSC; 2025. [Google Scholar]
32. European Centre for Disease Prevention and Control . European Standards of HIV Prevention and Care: Module on HIV Testing. ECDC; 2025. [Google Scholar]
33. Reynolds B, Naughton R, Lynn N, et al. HIV testing in primary care in the West of Ireland: a mixed‐method survey between 2013 and 2023. BJGP Open. 2025;9:BJGPO.2025.0034. [DOI] [PMC free article] [PubMed] [Google Scholar]
34. Gokul J, Chan KY, Dahab T, et al. HIV screening in acute medical patients presenting with an indicator condition. Ir Med J. 2025;118(9):148. [PubMed] [Google Scholar]
35. Abrantes R, Pimentel V, Miranda MNS, et al. Determinants of HIV late presentation among men who have sex with men in Portugal (2014‐2019): who's being left behind? Front Public Health. 2024;12:1336845. [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Brännström J, Svedhem Johansson V, Marrone G, et al. Deficiencies in the health care system contribute to a high rate of late HIV diagnosis in Sweden. HIV Med. 2016;17(6):425‐435. [DOI] [PubMed] [Google Scholar]
37. Lauscher P, Hanhoff N, Valbert F, et al. Socio‐demographic and psycho‐social determinants of HIV late presentation in Germany–results from the FindHIV study. AIDS Care. 2023;35(11):1749‐1759. [DOI] [PubMed] [Google Scholar]
38. Te Paske KS, Brand JM, Delsing C, et al. Trends and associated factors of late‐stage HIV diagnosis among heterosexuals in The Netherlands between 2012 and 2022: an analysis of national surveillance data. Int J Infect Dis. 2026;164:108324. [DOI] [PubMed] [Google Scholar]
39. Youssef E, Cooper V, Delpech V, Davies K, Wright J. Barriers and facilitators to HIV testing in people age 50 and above: a systematic review. Clin Med. 2017;17(6):508‐520. [DOI] [PMC free article] [PubMed] [Google Scholar]
40. Pantke A, Hoebel J, An Der Heiden M, et al. The impact of regional socioeconomic deprivation on the timing of HIV diagnosis: a cross‐sectional study in Germany. BMC Infect Dis. 2022;22(1):258. [DOI] [PMC free article] [PubMed] [Google Scholar]
41. Endalamaw A, Gilks CF, Ambaw F, Shiferaw WS, Assefa Y. Explaining inequity in knowledge, attitude, and services related to HIV/AIDS: a systematic review. BMC Public Health. 2024;24(1):1815. [DOI] [PMC free article] [PubMed] [Google Scholar]
42. Hachfeld A, Ledergerber B, Darling K, et al. Reasons for late presentation to HIV care in Switzerland. J Int AIDS Soc. 2015;18(1):20317. [DOI] [PMC free article] [PubMed] [Google Scholar]
43. Krankowska DC, Lourida P, Quirke SM, Woode Owusu M, Weis N, EACS/WAVE Migrant Women Working Group . Barriers to HIV testing and possible interventions to improve access to HIV healthcare among migrants, with a focus on migrant women: results from a European survey. HIV Med. 2024;25(5):554‐564. [DOI] [PubMed] [Google Scholar]
44. European Centre for Disease Prevention and Control . HIV and Migrants in the EU/EEA. Monitoring Implementation of the Dublin Declaration on Partnership to Fight HIV/AIDS in Europe and Central Asia: 2024 Progress Report. ECDC; 2024. [Google Scholar]
45. European Centre for Disease Prevention and Control . Public Health Guidance on Screening and Vaccination for Infectious Diseases in Newly Arrived Migrants Within the EU/EEA. ECDC; 2018. [Google Scholar]
46. Fuyama K, Hagiwara Y, Matsuyama Y. A simulation study of regression approaches for estimating risk ratios in the presence of multiple confounders. Emerg Themes Epidemiol. 2021;18(1):18. [DOI] [PMC free article] [PubMed] [Google Scholar]
47. Rose C, Pan Y, Baughman A. Bayesian logistic regression modeling as a flexible alternative for estimating adjusted risk ratios in studies with common outcomes. J Biom Biostat. 2015;6(4):1000253. [Google Scholar]
48. Charniga K, Park SW, Akhmetzhanov AR, et al. Best practices for estimating and reporting epidemiological delay distributions of infectious diseases. PLoS Comput Biol. 2024;20(10):e1012520. [DOI] [PMC free article] [PubMed] [Google Scholar]
49. Deblonde J, De Koker P, Hamers FF, Fontaine J, Luchters S, Temmerman M. Barriers to HIV testing in Europe: a systematic review. Eur J Public Health. 2010;20(4):422‐432. [DOI] [PubMed] [Google Scholar]
50. Stirrup O, Dunn D. Estimation of delay to diagnosis and incidence in HIV using indirect evidence of infection dates. BMC Med Res Methodol. 2018;18(1):65. [DOI] [PMC free article] [PubMed] [Google Scholar]
51. Stirrup O, Tostevin A, Ragonnet‐Cronin M, et al. Diagnosis delays in the UK according to pre or postmigration acquisition of HIV. AIDS. 2022;36(3):415‐422. [DOI] [PMC free article] [PubMed] [Google Scholar]
52. Rosinska M, Pantazis N, Janiec J, et al. Potential adjustment methodology for missing data and reporting delay in the HIV Surveillance System, European Union/European Economic Area, 2015. Euro Surveill. 2018;23(23):1700359. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure S1. Missingness pattern among new HIV diagnoses, Ireland 2015–2024.

Figure S2. Posterior predictive check for the primary logistic regression model showing observed and predicted late diagnosis proportion by covariate groups.

Figure S3. Posterior predictive check for the primary logistic regression model showing binned error plot.

Figure S8. Comparison of imputed versus observed CD4 count distributions by new diagnosis status. Results from 10 imputed datasets shown.

Figure S9. Comparison of imputed versus observed distributions for categorical variables, among those with new diagnoses. Results from 10 imputed datasets shown.

HIV-27-1881-s001.docx^{(736.6KB, docx)}

Data Availability Statement

[hiv70311-bib-0001] 1. Croxford S, Stengaard AR, Brännström J, et al. Late diagnosis of HIV: an updated consensus definition. HIV Med. 2022;23(11):1202‐1208. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0002] 2. Guaraldi G, Zona S, Menozzi M, et al. Late presentation increases risk and costs of non‐infectious comorbidities in people with HIV: an Italian cost impact study. AIDS Res Ther. 2017;14(1):8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0003] 3. Late Presentation Working Groups in EuroSIDA and COHERE . Estimating the burden of HIV late presentation and its attributable morbidity and mortality across Europe 2010–2016. BMC Infect Dis. 2020;20(1):728. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0004] 4. Mocroft A, Lundgren JD, Sabin ML, et al. Risk factors and outcomes for late presentation for HIV‐positive persons in Europe: results from the Collaboration of Observational HIV Epidemiological Research Europe Study (COHERE). PLoS Med. 2013;10(9):e1001510. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0005] 5. Reyes‐Urueña J, Stoppa G, Pizzolato F, Marrone G, Hansson D, EU/EEA HIV Network . Late HIV diagnosis: trends, risk factors, and progress toward the 2025 target of <20% late diagnosis in 23 EU/EEA countries, 2022 to 2024. Euro Surveill. 2025;30(47):2500855. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0006] 6. World Health Organization Regional Office for Europe (WHO/Europe) . Regional Action Plans for Ending AIDS and the Epidemics of Viral Hepatitis and Sexually Transmitted Infections, 2022–2030. WHO/Europe; 2023. [Google Scholar]

[hiv70311-bib-0007] 7. Greenland S. Noncollapsibility, confounding, and sparse‐data bias. Part 1: the oddities of odds. J Clin Epidemiol. 2021;138:178‐181. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0008] 8. McNutt L‐A. Estimating the relative risk in cohort studies and clinical trials of common outcomes. Am J Epidemiol. 2003;157(10):940‐943. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0009] 9. Greenland S, Maldonado G. The interpretation of multiplicative‐model parameters as standardized parameters. Stat Med. 1994;13(10):989‐999. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0010] 10. Zou G. A modified Poisson regression approach to prospective studies with binary data. Am J Epidemiol. 2004;159(7):702‐706. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0011] 11. Muller CJ, MacLehose RF. Estimating predicted probabilities from logistic regression: different methods correspond to different target populations. Int J Epidemiol. 2014;43(3):962‐970. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0012] 12. Xia Q, Kobrak P, Wiewel EW, Torian LV. The high proportion of late HIV diagnoses in the USA is likely to stay: findings from a mathematical model. AIDS Care. 2015;27(2):206‐212. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0013] 13. HSE – Health Protection Surveillance Centre . Notifiable Diseases. HPSC; 2025. https://www.hpsc.ie/p/notifiablediseases/listofnotifiablediseases. Accessed February 2, 2026. [Google Scholar]

[hiv70311-bib-0014] 14. Bürkner PC. Bayesian item response modeling in R with brms and Stan. J Stat Softw. 2021;100(5):1‐54. [Google Scholar]

[hiv70311-bib-0015] 15. Department of Health . HSE Health Regions. Government of Ireland; 2025. https://www.gov.ie/en/department-of-health/publications/hse-health-regions/. Accessed February 2, 2026. [Google Scholar]

[hiv70311-bib-0016] 16. McHenry H, Frost D. WDC Census 2022 Summary Report – Highlights for Western Region. Western Development Commission Policy Analysis Team; 2023. [Google Scholar]

[hiv70311-bib-0017] 17. Vehtari A, Gabry J, Magnusson M, et al. loo: efficient leave‐one‐out cross‐validation and WAIC for bayesian models. 2024. https://mc-stan.org/loo/. Accessed February 2, 2026.

[hiv70311-bib-0018] 18. Pantazis N, Thomadakis C, del Amo J, et al. Determining the likely place of HIV acquisition for migrants in Europe combining subject‐specific information and biomarkers data. Stat Methods Med Res. 2019;28(7):1979‐1997. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0019] 19. Pantazis N, Rosinska M, van Sighem A, et al. Discriminating between premigration and postmigration HIV acquisition using surveillance data. J Acquir Immune Defic Syndr. 2021;88(2):117‐124. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0020] 20. Van Sighem A, Pharris A, Quinten C, Noori T, Amato‐Gauci AJ, ECDC HIV/AIDS Surveillance and Dublin Declaration Monitoring Networks . Reduction in undiagnosed HIV infection in the European Union/European Economic Area, 2012 to 2016. Euro Surveill. 2017;22(48):17‐00771. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0021] 21. Gbadamosi SO, Trepka MJ, Dawit R, Jebai R, Sheehan DM. A systematic review and meta‐analysis to estimate the time from HIV infection to diagnosis for people with HIV. AIDS Rev. 2022;24(1):32‐40. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0022] 22. van Buuren S, Groothuis‐Oudshoorn K. Mice: multivariate imputation by chained equations in r. J Stat Softw. 2011;45(3):1‐67. [Google Scholar]

[hiv70311-bib-0023] 23. Bartlett J, Keogh R, Bonneville EF, van der Burg L. smcfcs: Multiple imputation of covariates by substantive model compatible fully conditional specification. 2025, https://CRAN.R-project.org/package=smcfcs. Accessed February 2, 2026.

[hiv70311-bib-0024] 24. R Core Team . R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2025. [Google Scholar]

[hiv70311-bib-0025] 25. Jasperse J, Harris RJ, Humphreys C, et al. Risk factors for late HIV diagnosis in England, 2015–2023. HIV Med. 2025;27(3):409‐419. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0026] 26. Department of Health . The National Sexual Health Strategy 2025–2035. Government of Ireland; 2025. [Google Scholar]

[hiv70311-bib-0027] 27. Bath RE, Emmett L, Verlander NQ, Reacher M. Risk factors for late HIV diagnosis in the East of England: evidence from national surveillance data and policy implications. Int J STD AIDS. 2019;30(1):37‐44. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0028] 28. Noble G, Okpo E, Tonna I, Fielding S. Factors associated with late HIV diagnosis in North‐East Scotland: a six‐year retrospective study. Public Health. 2016;139:36‐43. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0029] 29. Op De Coul ELM, Van Sighem A, Brinkman K, et al. Factors associated with presenting late or with advanced HIV disease in the Netherlands, 1996–2014: results from a national observational cohort. BMJ Open. 2016;6(1):e009688. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0030] 30. Parczewski M, Gökengin D, Sullivan A, et al. Control of HIV across the WHO European region: progress and remaining challenges. Lancet Reg Health Eur. 2025;52:101243. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0031] 31. HSE – Health Protection Surveillance Centre . HIV Slideset 2024. HPSC; 2025. [Google Scholar]

[hiv70311-bib-0032] 32. European Centre for Disease Prevention and Control . European Standards of HIV Prevention and Care: Module on HIV Testing. ECDC; 2025. [Google Scholar]

[hiv70311-bib-0033] 33. Reynolds B, Naughton R, Lynn N, et al. HIV testing in primary care in the West of Ireland: a mixed‐method survey between 2013 and 2023. BJGP Open. 2025;9:BJGPO.2025.0034. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0034] 34. Gokul J, Chan KY, Dahab T, et al. HIV screening in acute medical patients presenting with an indicator condition. Ir Med J. 2025;118(9):148. [PubMed] [Google Scholar]

[hiv70311-bib-0035] 35. Abrantes R, Pimentel V, Miranda MNS, et al. Determinants of HIV late presentation among men who have sex with men in Portugal (2014‐2019): who's being left behind? Front Public Health. 2024;12:1336845. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0036] 36. Brännström J, Svedhem Johansson V, Marrone G, et al. Deficiencies in the health care system contribute to a high rate of late HIV diagnosis in Sweden. HIV Med. 2016;17(6):425‐435. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0037] 37. Lauscher P, Hanhoff N, Valbert F, et al. Socio‐demographic and psycho‐social determinants of HIV late presentation in Germany–results from the FindHIV study. AIDS Care. 2023;35(11):1749‐1759. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0038] 38. Te Paske KS, Brand JM, Delsing C, et al. Trends and associated factors of late‐stage HIV diagnosis among heterosexuals in The Netherlands between 2012 and 2022: an analysis of national surveillance data. Int J Infect Dis. 2026;164:108324. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0039] 39. Youssef E, Cooper V, Delpech V, Davies K, Wright J. Barriers and facilitators to HIV testing in people age 50 and above: a systematic review. Clin Med. 2017;17(6):508‐520. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0040] 40. Pantke A, Hoebel J, An Der Heiden M, et al. The impact of regional socioeconomic deprivation on the timing of HIV diagnosis: a cross‐sectional study in Germany. BMC Infect Dis. 2022;22(1):258. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0041] 41. Endalamaw A, Gilks CF, Ambaw F, Shiferaw WS, Assefa Y. Explaining inequity in knowledge, attitude, and services related to HIV/AIDS: a systematic review. BMC Public Health. 2024;24(1):1815. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0042] 42. Hachfeld A, Ledergerber B, Darling K, et al. Reasons for late presentation to HIV care in Switzerland. J Int AIDS Soc. 2015;18(1):20317. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0043] 43. Krankowska DC, Lourida P, Quirke SM, Woode Owusu M, Weis N, EACS/WAVE Migrant Women Working Group . Barriers to HIV testing and possible interventions to improve access to HIV healthcare among migrants, with a focus on migrant women: results from a European survey. HIV Med. 2024;25(5):554‐564. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0044] 44. European Centre for Disease Prevention and Control . HIV and Migrants in the EU/EEA. Monitoring Implementation of the Dublin Declaration on Partnership to Fight HIV/AIDS in Europe and Central Asia: 2024 Progress Report. ECDC; 2024. [Google Scholar]

[hiv70311-bib-0045] 45. European Centre for Disease Prevention and Control . Public Health Guidance on Screening and Vaccination for Infectious Diseases in Newly Arrived Migrants Within the EU/EEA. ECDC; 2018. [Google Scholar]

[hiv70311-bib-0046] 46. Fuyama K, Hagiwara Y, Matsuyama Y. A simulation study of regression approaches for estimating risk ratios in the presence of multiple confounders. Emerg Themes Epidemiol. 2021;18(1):18. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0047] 47. Rose C, Pan Y, Baughman A. Bayesian logistic regression modeling as a flexible alternative for estimating adjusted risk ratios in studies with common outcomes. J Biom Biostat. 2015;6(4):1000253. [Google Scholar]

[hiv70311-bib-0048] 48. Charniga K, Park SW, Akhmetzhanov AR, et al. Best practices for estimating and reporting epidemiological delay distributions of infectious diseases. PLoS Comput Biol. 2024;20(10):e1012520. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0049] 49. Deblonde J, De Koker P, Hamers FF, Fontaine J, Luchters S, Temmerman M. Barriers to HIV testing in Europe: a systematic review. Eur J Public Health. 2010;20(4):422‐432. [DOI] [PubMed] [Google Scholar]

[hiv70311-bib-0050] 50. Stirrup O, Dunn D. Estimation of delay to diagnosis and incidence in HIV using indirect evidence of infection dates. BMC Med Res Methodol. 2018;18(1):65. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0051] 51. Stirrup O, Tostevin A, Ragonnet‐Cronin M, et al. Diagnosis delays in the UK according to pre or postmigration acquisition of HIV. AIDS. 2022;36(3):415‐422. [DOI] [PMC free article] [PubMed] [Google Scholar]

[hiv70311-bib-0052] 52. Rosinska M, Pantazis N, Janiec J, et al. Potential adjustment methodology for missing data and reporting delay in the HIV Surveillance System, European Union/European Economic Area, 2015. Euro Surveill. 2018;23(23):1700359. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Late HIV diagnosis: An application of flexible Bayesian approaches to understand associated factors and time to diagnosis in Ireland, 2015–2024

Cian Dowling‐Cullen

Katie O'Brien

Kate O'Donnell

Mary Archibald

Phil Downes

David Field

Caroline Hurley

Fiona Lyons

Eavan Muldoon

Peter Barrett

Cathal Walsh

Derval Igoe

Abstract

Objectives

Methods

Results

Conclusion

INTRODUCTION

METHODS

Data source and study population

Definitions

Statistical analysis

Logistic regression with marginal standardization

Time‐to‐diagnosis modelling

Missing data handling

Sensitivity analyses

Missing data

Diagnosis date

Probable country of infection

Delay analysis priors and distribution

RESULTS

Descriptive results

FIGURE 1.

TABLE 1.

FIGURE 2.

Logistic regression

TABLE 2.

Time‐to‐diagnosis distributions

TABLE 3.

TABLE 4.

Sensitivity analyses

DISCUSSION

Key findings and comparison with other evidence

Strengths and limitations

CONCLUSION

AUTHOR CONTRIBUTIONS

CONFLICT OF INTEREST STATEMENT

ETHICS STATEMENT

Supporting information

ACKNOWLEDGEMENTS

DATA AVAILABILITY STATEMENT

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases