Factors associated with the level of CD4 cell counts at HIV diagnosis in a French cohort: a quantile regression approach

Abstract

The consensus definition of late presentation for human immunodeficiency virus patient based on a CD4 threshold of 350 cells/mm³ has limitations concerning risk factors identification since there is growing biomedical justification for earlier initiation of treatment. The objective was to overcome this problem by simultaneously determining factors associated with different levels of CD4 counts at the time of diagnosis. Between January 2000 and July 2014, 1179 patients with a first human immunodeficiency virus diagnosis and entering care in a French human immunodeficiency virus reference center were enrolled. Factors associated with each 5 percentile from 5th to 95th quantile of CD4 counts at diagnosis were simultaneously studied in a multivariable quantile regression model. At each of the quantiles, the factors identified as negatively associated with CD4 count at diagnosis were older age, male sex , foreign patients, hepatitis B virus or hepatitis C virus co-infection, employment status, non-MSM transmission, heterosexual transmission, suburban and rural’s place of residence and earlier period of diagnosis. Association with CD4 count was not uniformly significant, most factors being significant for some quantiles. The only significant determinant for all quantiles was being born in a foreign country. These results are particularly helpful in the context of human immunodeficiency virus clinical care, management and prevention.

Keywords

HIV/AIDS late diagnosis CD4 counts France quantile regression

Introduction

The human immunodeficiency virus (HIV) infection is characterized by a progressive loss of CD4 cells, resulting in immunosuppression leading to the development of opportunistic infections.¹ The rate at which CD4 cell counts decline is one of the main markers of HIV disease progression and is a widely used parameter to guide antiretroviral therapy (ART) initiation as well as to assess its efficacy. Despite the medical and societal benefits of an early diagnosis,² a large number of HIV-positive persons continue to be diagnosed with HIV at a late stage of infection in developed countries.³

In 2011, a consensus definition of late presentation (LP) was approved by the European Consensus panel.⁴ It was defined as persons presenting for care with a CD4 count below 350 cells/µL or presenting with an AIDS-defining event, regardless of the CD4 cell count. This definition is particularly useful to describe the level of disease progression and allows for a consistent evaluation of associated determinants. The identification of LP determinants is critical in adapting and designing more effective strategies and prevention campaign.⁴

The majority of recent studies on this topic have thus relied on this consensus definition.^3,5,6 As noted elsewhere,⁷ the definition of LP based on a CD4 threshold of 350 cells/µL has important limitations since there is growing biomedical justification for earlier initiation of treatment. In the international IAS-USA guidelines, since 2012, ART is recommended in all persons living with HIV regardless of their CD4 counts, even if it is > 500 cells/µL.¹

In this context, the use of a particular CD4 cut-off may not be the more pertinent method to identify risk factors associated with LP.⁸ Indeed, using such a predefined cut-off is problematic since the potential risk factors may be differently associated with CD4 counts at various points of its distribution. For example, some factors might be weakly associated with average CD4 counts, whereas it might be strongly associated with lower CD4 counts. Understanding such differential associations is nevertheless crucial to identify the different populations at risk according to different levels of CD4 counts at HIV diagnosis. To do so, the use of innovative statistical procedures such as quantile regression⁹ may allow a more comprehensive analysis of factors associated with CD4 count in comparison to standard logistic regression. It has also the supplementary advantage to quantify the relative loss of CD4 cells associated with a particular risk factor.¹⁰

The aim of this study was thus to alternatively use quantile regression to account for potential heterogeneity in the association of various explanatory variables across different quantiles of CD4 counts at the time of diagnosis in HIV patients followed in a French University Hospital Center between 2000 and 2014.

Material and methods

Data from the Nantes University Hospital Center NADIS cohort were used.¹¹ NADIS is an electronic medical record for HIV, hepatitis B virus (HBV) or hepatitis C virus (HCV)-infected adults seeking care in French public hospitals. All patients’ data are prospectively recorded on a real-time basis in a structured database, allowing use of the database for clinical, epidemiological or therapeutic studies. In Nantes, the largest city in the northwest of France, Nadis medical software has been used in the HIV reference center of the University Hospital since 2000. In this study, all HIV-positive adult patients with a first HIV diagnosis between January 2000 and July 2014 were enrolled. Included patients gave informed consent to the use of their data. The data collection was approved by the National Competent Authority.

The primary endpoint was initial CD4 counts (per µL) recorded in Nadis within six months after the first positive HIV serology. This variable was kept as continuous. Investigated factors were sex, age at diagnosis, year of HIV diagnosis, mode of transmission, HBV or HCV co-infection, country of birth, employment status, relationship status, and urbanicity of place of residence as defined by the National Institute for Statistics and Economic Studies.¹²

A multivariate quantile regression¹³ model was used to test the association between CD4 counts and suspected determinants. Traditional regression models assume that the impact of the independent variables is the same at different quantiles of the CD4 distribution. Quantile regression has the advantage to provide a more exact description of the distribution of a conditional variable in its determinants than traditional regression models.¹³ The quantile model was thus chosen to better understand whether the determinants of CD4 counts at diagnosis change at different levels of the CD4 distribution. The coefficients were estimated for each 5 percentile from 5th to 95th quantile of CD4 count. Standard errors and 95% confidence intervals (CI) were estimated using bootstrapping with 1000 replications. The coefficients for each quantile were plotted as well as their 95% CI. All statistical analyses were performed using R software.¹⁴

Results

Between January 2000 and July 2014, 1179 patients met inclusion criteria with no missing data concerning CD4 counts in six months following diagnosis. Their characteristics are summarized in Table 1. The median CD4 cell count at diagnosis was 366/µL (interquartile range [IQR]: 206–558). The 10th quantile of CD4 cell count was 54 CD4/µL. The 10th quantile means that 10% of included patients had CD4 counts at diagnosis less than 54/µL, and 90% of patients had CD4 counts at diagnosis more than 54/µL. The median age at diagnosis was 35 years (IQR: 29–44). The majority was men (68%). Forty-eight percent were employed, 44% were single or living apart, 56% were French citizen and 70% lived in urban areas. The estimated mode of infection was heterosexual for 47% of the patients. Nine percent were co-infected with HBV (5%) or HCV (4%). Results of the quantile regression analysis are shown in Figure 1.

Table 1.

Demographic and clinical characteristics of HIV-positive adult patients with a first diagnosis in the HIV reference center of the University Hospital between January 2000 and July 2014 (n = 1179).

Characteristics			CD4 count at diagnosis/µL
Characteristics		n (%)	Median	IQR 25–75
All patients		1179 (100)	366	206–558
Age at diagnosis	<30 years	373 (32)	432	288–619
	30–39 years	378 (32)	360	188–539
	40–49 years	272 (23)	336	170–551
	≥50 years	156 (13)	256	98–472
Country of birth	France	663 (56)	418	222–601
	Foreign	392 (33)	312	182–464
	Unknown	124 (11)	373	228–553
Employment status	Employed	562 (48)	425	242–586
	Unemployed	237 (20)	331	186–556
	Other^a	110 (9)	300	158–547
	Unknown	270 (23)	320	164–496
Sex	Male	800 (68)	394	197–562
Sex	Female	379 (32)	334	215–539
HIV transmission category	Heterosexual	559 (47)	335	190–530
	MSM^b	470 (40)	440	253–608
	Others	150 (13)	291	121–429
Year of HIV diagnosis	2000–2003	307 (26)	334	154–560
	2004–2006	283 (24)	357	219–543
	2007–2010	327 (28)	389	212–550
	2011–2014	262 (22)	396	220–592
HBV or HCV co-infection	No	1067 (91)	372	215–560
	HBV	66 (5)	247	82–427
	HCV	46 (4)	386	202–630
Relationship status	Couple	425 (36)	336	208–552
	Single	523 (44)	401	211–572
	Unknown	231 (20)	362	190–544
Place of residence	Urban areas	829 (70)	389	225–564
	Rural areas	141 (12)	342	119–572
	Suburban areas	209 (18)	314	138–511

IQR: interquartile range; HBV: hepatitis B virus; HCV: hepatitis C virus.

student, retired, pensioner.

Men who have sex with men.

Figure 1.

Multivariate quantile regression between CD4 count at first diagnosis and suspected determinants in patients followed in the HIV reference center of the University Hospital between January 2000 and July 2014 (n = 1179). Note: The x-axis represents quantiles of the distribution of CD4 count, and the y-axis represents the change in CD4 count associated with a one-unit change in the studied covariate, holding other covariates constant. This change is considered significant at a particular quantile when the associated 95% confidence interval (shaded area) does not cross the 0 value (dot line). **The intercept can be interpreted as the estimates of CD4 count at diagnosis across the studied quantiles given all covariates set to zero (i.e. CD4 count at diagnosis for French male patient aged 18 years old, diagnosed in 2000, living in an urban area, being employed, MSM as a mode of HIV transmission, with no HBV or HCV co-infection and being alone).

Female gender was positively associated with CD4 count at diagnosis only for lower quantiles (≤20th quantile) with the female to male difference in CD4 counts varying from 32 (95%CI: 10–53) to 66 (95%CI: 38–93) CD4/µL. The associations of age at diagnosis on CD4 count were significant for most quantiles (≤80th quantile) and more pronounced for intermediary ones. Higher age was associated with significantly lower CD4 count at diagnosis. At the 50th quantile, an increase of 10 years in age was associated to a decrease of 44 (95%CI: 27–61) CD4/µL at diagnosis. People infected by heterosexual or non-sexual transmission had significantly lower CD4 counts than men who have sex with men (MSM) for quantiles respectively ≤60th and ≤80th quantile. These associations were fairly constant across the quantiles but more pronounced for people infected by non-sexual transmission. The significance of an association between year of diagnosis and CD4 count at diagnosis was consistent only for quantiles ≤45th quantile. At the 20th quantile, an increase of 10 years was associated with an increase of 64 (95%CI: 36–93) CD4/µL at diagnosis. Being foreign born was inversely related with CD4 count at diagnosis. The magnitude of the association displayed a decreasing trend from low to high quantiles. At the 50th quantile, being foreign born was associated with a decrease of 78 (95%CI: 31–124) CD4/µL compared to French patients.

HBV or HCV co-infection was only negatively linked with CD4 count at diagnosis for lower quantiles (≤20th quantile). For these quantiles, effects on CD4 count at diagnosis oscillated between 23 (95%CI: 7–40) and 59 (95%CI: 2–117) CD4/µL. Unemployed patients had significantly lower CD4 cell counts at diagnosis compared to employed people only for intermediate quantiles (between 35th and 50th quantiles). At the 50th quantile, being unemployed was associated with a decrease of 61 (95%CI: 11–111) CD4/µL compared to employed patients. Patients living in suburban and rural areas compared to their urban counterparts had significantly lower rates of CD4 at diagnosis for low and intermediate quantiles (≤60th quantile). This association was more pronounced for suburban areas than for rural ones as well as for intermediary quantiles. Being in a couple did not present consistent relations with CD4 count at diagnosis whatever the studied quantiles.

Discussion

The CD4 threshold for initiating ART has been constantly increasing over the last years leading patients diagnosed late today to be different from patients late diagnosed a few years ago. To overcome the problem of defining an arbitrary threshold to study factors associated with CD4 count at diagnosis, we used quantile regression. This statistical method allows to simultaneously determine the associations of suspected determinants on various quantiles of CD4 counts. It also has the supplementary advantage to quantify the relative loss of CD4 cells associated with a particular risk factor.^8–10 Our results demonstrated that quantile regression provide a more comprehensive analysis of the data than traditional logistic regression. Indeed, this method has enabled us to develop a better understanding of important factors associated with late HIV diagnosis. Despite its potential advantages, the utilization of quantile regression remains underutilized in the area. We found only one study using quantile regression to identify determinants of LP.¹⁵ In this last, quantile regression was only used to examine changes in CD4 cell count at presentation and trends over calendar time.

At each of the quantiles, the factors identified as negatively associated with CD4 count at diagnosis were consistent with those identified in previous studies included older age,^3,6,16–18 male sex,^3,6,15 foreign patients,^3,17,18 HBV or HCV co-infection,^19,20 employment status,²⁰ non-MSM transmission,^3,16 heterosexual transmission^6,18 and older period of diagnosis.^3,6 The factors associated with CD4 level were significant at levels of CD4 varying between 30 and 200 cells, indicating that they were not negligible from a clinical perspective. There are no studies conducted in France or in Europe analyzing urban versus rural residence on LP and/or CD4 count at HIV diagnosis. However, like the current study, a recent American study similarly showed that rural people were more likely to be diagnosed late than urban people.²¹ Our result thus suggests that screening in France have to be developed outside the largest cities, particularly in suburban areas, to avoid late diagnosis.

The present study showed heterogeneity of coefficients for studied determinants and adds to the literature by demonstrating that some of them have only negative effect for the lower percentiles (≤20th percentile, i.e. for CD4 count at diagnosis ≤156 CD4/µL), the intermediate quantiles (roughly comprised between the 20th and the 70th quantile, i.e. for CD4 count between 156 and 524 CD4/µL) or for both. The only significant determinant for all quantiles was being born in a foreign country. The magnitude of its association with CD4 count at diagnosis was characterized by an increasing trend for low to high quantiles. This result underlines that dedicated actions aiming to diminish structural, healthcare and community barriers to HIV testing among migrants are necessary in many countries, including France.²²

With the growing biomedical justification for earlier initiation of treatment, these results are particularly helpful in the context of HIV clinical care, management and prevention. Even if current recommendation is for universal ART, independently of CD4 cell counts at presentation, the delay in treatment initiation can be adapted to the level of CD4. Patients with CD4 around or above 500/mm³ are not at immediate risk of clinical progression, while those with CD4 <200/mm³ need immediate intervention, and those with intermediate CD4 should be rapidly initiated on treatment as CD4 and CD4:CD8 ratio complete recovery is hampered in patients with CD4 below 500/mm³, as a function of CD4 decrease. Furthermore, long-term comorbities and non-AIDS events are associated to nadir CD4. Therefore, although from a strict ART indication, knowing the CD4 cell counts ranges at diagnosis is not useful, this remains highly relevant for clinical care and prevention strategies. Indeed, the quantile methods provide insights on the repartition of CD4 in different populations at diagnosis. The results can be used to identify people who are at higher risk for varying degrees of late diagnosis and thus aid in prioritization and identification of groups that may need treatment urgently. From a prevention standpoint, information on repartition of ranges of CD4 at late diagnosis is also very useful to target specific actions to improve early diagnosis and avoid missed opportunities in the different risk groups.

The main limitation of our study is the relative small sample size and its associated potential lack of power. This might have led us to fail to detect important trends in our data. Our results are, however, consistent with the literature and deserve consideration for a larger scale study. Another limitation was that we excluded people without CD4 in six months following diagnosis potentially leading to selection bias. Moreover, only CD4 counts and not the presence of AIDS defining event was considered in our study, which focused on late diagnosis. Late diagnosis and LP are often mistakenly used in the literature to identify patients presenting late in care, and the breadth and inconsistency of these labels have been largely discussed.²³ Late diagnosis reflects the CD4 cell count closest to first HIV diagnosis test. Late presentation reflects the immunologic (CD4 count) and clinical status at the first HIV medical care visit. The first one is more relevant from an epidemiological standpoint. The second one is more relevant clinically, but could miss relevant information at time of diagnosis, if there is a long delay for first HIV clinic visit and/or absence of CD4 testing after first HIV test is found positive.

In summary, this study provides new information to the current knowledge of risk factors associated with late diagnosis, with a focus on French population. Its methodological novelty creates some new directions for future research, and sets the stage for a line of national and international research using the quantile regression approach.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Ethical approval

Written consent was obtained for each patient before inclusion in the study, and the cohort was registered at the French data protection authority in clinical research (Commission Nationale de l’Informatique et des Libertés or CNIL, no. 851117).

Contributorship

FR and MH contributed equally to this article. LB: drafted the initial manuscript, carried out the initial analyses and approved the final manuscript as submitted. MH: conceptualized and designed the study, carried out the initial analyses, provided guidance on data interpretation, reviewed and revised the manuscript, and approved the final manuscript as submitted. EB: approved the final manuscript as submitted. FR: provided guidance on data interpretation, reviewed and revised the manuscript, and approved the final manuscript as submitted.

References

Günthard

Aberg

Eron

. Antiretroviral treatment of adult HIV infection: 2014 recommendations of the International Antiviral Society-USA Panel. JAMA 2014; 312: 410–425.

Moreno

Mocroft

Monforte

A d’Arminio

. Medical and societal consequences of late presentation. Antivir Ther 2010; 15(Suppl 1): 9–15.

Mocroft

Lundgren

Sabin

. 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: e1001510–e1001510.

Antinori

Coenen

Costagliola

. Late presentation of HIV infection: a consensus definition. HIV Med 2011; 12: 61–64.

Zoufaly

an der Heiden

Marcus

. Late presentation for HIV diagnosis and care in Germany. HIV Med 2012; 13: 172–181.

Celesia

Castronuovo

Pinzone

. Late presentation of HIV infection: predictors of delayed diagnosis and survival in Eastern Sicily. Eur Rev Med Pharmacol Sci 2013; 17: 2218–2224.

MacCarthy

Bangsberg

Fink

. Late presentation to HIV/AIDS testing, treatment or continued care: clarifying the use of CD4 evaluation in the consensus definition. HIV Med 2014; 15: 130–134.

Bennette

Vickers

. Against quantiles: categorization of continuous variables in epidemiologic research, and its discontents. BMC Med Res Methodol 2012; 12: 21–21.

Beyerlein

. Quantile regression – opportunities and challenges from a user’s perspective. Am J Epidemiol 2014; 180: 330–331.

10.

Rehkopf

. Quantile regression for hypothesis testing and hypothesis screening at the dawn of big data. Epidemiology 2012; 23: 665–667.

11.

Pugliese

Cuzin

Cabié

. A large French prospective cohort of HIV-infected patients: the Nadis Cohort. HIV Med 2009; 10: 504–511.

12.

National Institute of Statistics and Economic Studies Urban unit, www.insee.fr/en/methodes/default.asp?page=definitions/unite-urbaine.htm (2014, accessed 5 May 2016).

13.

Koenker

Hallock

. Quantile regression. J Econ Perspect 2001; 15: 143–156.

14.

R Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, www.R-project.org/ (2013, accessed 5 May 2016).

15.

Borghi

Girardi

Bellelli

. Late presenters in an HIV surveillance system in Italy during the period 1992-2006. J Acquir Immune Defic Syndr 1999; 49: 282–286.

16.

Wilson

K d’Almeida

Dray-Spira

Aubrière

. Frequency and correlates of late presentation for HIV infection in France: older adults are a risk group – results from the ANRS-VESPA2 Study, France. AIDS Care 2014; 26(Suppl 1): S83–S93.

17.

Buetikofer

Wandeler

Kouyos

. Prevalence and risk factors of late presentation for HIV diagnosis and care in a tertiary referral centre in Switzerland. Swiss Med Wkly 2014; 144: w13961–w13961.

18.

Yombi

Jonckheere

Vincent

. Late presentation for human immunodeficiency virus HIV diagnosis results of a Belgian single centre. Acta Clin Belg 2014; 69: 33–39.

19.

Agaba

Meloni

Sule

. Patients who present late to HIV care and associated risk factors in Nigeria. HIV Med 2014; 15: 396–405.

20.

Mannheimer

Wang

Wilton

. Infrequent HIV testing and late HIV diagnosis are common among a cohort of black men who have sex with men in 6 US cities. J Acquir Immune Defic Syndr 2014; 67: 438–445.

21.

Trepka

Fennie

Sheehan

. Late HIV diagnosis: differences by rural/urban residence, Florida, 2007-2011. AIDS Patient Care STDs 2014; 28: 188–197.

22.

Alvarez-del Arco

Monge

Caro-Murillo

. HIV testing policies for migrants and ethnic minorities in EU/EFTA member states. Eur J Public Health 2014; 24: 139–144.

23.

Kozak

Zinski

Leeper

. Late diagnosis, delayed presentation and late presentation in HIV: proposed definitions, methodological considerations and health implications. Antivir Ther 2013; 18: 17–23.