High rate of loss to clinical follow up among African HIV-infected patients attending a London clinic: a retrospective analysis of a clinical cohort

Background Long-term regular clinic follow up is an important component of HIV care. We determined the frequency and characteristics of HIV-infected patients lost to follow up from a London HIV clinic, and factors associated with loss to all HIV follow up in the UK. Methods We identified 1859 HIV-infected adults who had registered and attended a London clinic on one or more occasions between January 1997 and December 2005. Loss to follow up was defined as clinic non-attendance for one or more years. Through anonymized linkage with the Survey of Prevalent HIV Infections Diagnosed and Health Protection Scotland, national databases of all HIV patients in care in the UK up to December 2006, loss-to-follow-up patients were categorized as Transfers (subsequently received care at another UK HIV clinic) or UKLFU (no record of subsequent attendance at any HIV clinic in the UK). Logistic regression analysis was used to identify factors associated with UKLFU for those both on highly active antiretroviral therapy (HAART) and not on HAART. Results In total, 722 (38.8%) of 1859 patients were defined as lost to follow up. Of these, 347 (48.1%) were Transfers and 375 (51.9%), or 20.2% of all patients, were UKLFU. Overall, 11.9% of all patients receiving HAART, and 32.2% not receiving HAART were UKLFU. Among those on HAART, risk factors for UKLFU were: African heterosexual female (OR = 2.22, 95% CI: 1.11-4.56) versus white men who have sex with men; earlier year of HIV clinic registration (1997-1999 OR: 3.51, 95% CI: 1.97-6.26; 2000-02 OR: 2.49, 95% CI: 1.43-4.32 vs. 2003-2005); CD4 count of < 200 versus > 350 cells/mm3 (OR = 1.99, 95% CI:1.05-3.74); and a detectable viral load of > 400 copies/ml (OR = 5.03, 95% CI: 2.95-8.57 vs. ≤ 400 copies/ml) at last clinic visit. Among those not receiving HAART, factors were: African heterosexual male (OR = 3.91, 95% CI: 1.77-8.64) versus white men who have sex with men; earlier HIV clinic registration (2000-2002 OR: 2.91, 95% CI: 1.77-4.78; 1997-1999: OR: 5.26, 95% CI: 2.71-10.19); and a CD4 count of < 200 cells/mm3 (OR: 3.24, 95% CI: 1.49-7.04). Conclusions One in five HIV-infected patients (one in three not on HAART and one in nine on HAART) from a London clinic were lost to all clinical follow up in the UK. Black African ethnicity, earlier year of clinic registration and advanced immunological suppression were the most important predictors of UKLFU. There is a need for all HIV clinics to establish systems for monitoring and tracing loss-to-follow-up patients, and to implement strategies for improving retention in care.


Background
The widespread availability of highly active antiretroviral therapy (HAART) has transformed the prognosis of HIVinfected patients over the past 10 years [1,2]. However, while there has been a marked reduction in HIV-related mortality and morbidity [2,3], additional challenges have emerged in the long-term management of HIV, such as drug toxicities [4][5][6], treatment failure due to poor adherence [7,8] and/or drug resistance [9] requiring regimen change, and increasing non-HIV-related morbidity and mortality [10][11][12]. As with any chronic illness, HIV patients require long-term, regular clinical follow up to monitor disease progression and optimal timing of initiation of HAART, assess response and adherence to antiretroviral therapy and associated adverse events, and address sexual health and HIV prevention needs [13]. Long-term retention is therefore an important component of HIV care, and high rates of loss to follow up would compromise the effective delivery of HIV care, as well as reliable documentation of mortality.
Rates of adherence [7], virological suppression [14] and survival [14,15] are the most commonly reported measures of the effectiveness of HAART management, but this applies only to patients who remain under follow up and in care. Although there have been several studies on losses to follow up from HAART programmes across sub-Saharan Africa [16][17][18][19][20], until recently there had been a paucity of information on losses to follow up from HIV care in high-income countries [21][22][23][24][25][26]. Most of these studies were conducted prior to the widespread use of HAART [23,25,26] or in research cohorts [27][28][29]. Furthermore, since many clinics do not have established systems for identifying patients who are lost to follow up (LFU), there is need for a greater understanding of the rates and reasons for loss to follow up to develop strategies to optimize retention.
We undertook an evaluation of the frequency of LFU and characteristics of patients who were LFU among adult HIV-infected patients attending a large HIV clinic in south London. We then matched those patients who were LFU with the national databases of all HIV patients receiving care in the UK to establish whether patients were LFU due to transfer to another clinic in the UK, or were lost to all follow up in the UK, and the factors associated with this.

Study population
HIV care and treatment in the UK is delivered through 227 specialist treatment sites, of which King's College Hospital is the eighth largest clinic nationally and the second largest in south London (personal communication Chau Cuong, Survey of Prevalent HIV Infections Diagnosed, Health Protection Agency). The clinic is based in south-east London, which has the highest rate of new HIV diagnoses in the UK, accounting for approximately 10% of all new HIV diagnoses in 2006 [30]. The clinic population is also highly ethnically diverse -of 500 new HIV diagnoses between 1998 and 2000, 54.7% were in black Africans (34.8% from eastern Africa, 32.8% from west Africa, and 32.4% from southeast Africa) and 9.1% in black Caribbeans.

Identification and definition of loss to follow up
We identified all HIV-1 infected patients (≥ 18 years) who registered and attended the King's College Hospital HIV clinic on at least one occasion over a nine-year period (between 1 January 1997 and 31 December 2005), and reviewed their follow-up records up to 31 December 2006. The day, 1 January 1997, was identified as the initial date for the selection of the study population as by this date, HAART was widely available, and we wished to avoid the potential bias of availability of HAART on losses to follow up.
These patients were categorized as either LFU or current clinical attendees (CCAs). LFU was defined as nonattendance at the Kings College Hospital clinic for at least one year (up to 31 December 2006); at this clinic, routine follow up is in general every three months for patients on HAART, and every three to six months for those not yet on HAART. CCAs were defined as those patients who remained under King's College Hospital clinic follow up as of 31 December 2006.
Patients defined as LFU were then matched against two national registers of HIV patients receiving care in the UK: the Survey of Prevalent HIV Infections Diagnosed (SOPHID), which covers England, Wales and Northern Ireland; and the Health Protection Scotland database. The intention was to determine if these LFU patients had received care elsewhere in the UK between 1 January 1997 and 31 December 2006 (or 30 June 2007 for London SOPHID). Records were linked by matching on three criteria: soundex code of surname, sex and date of birth [31]. Additional matches were obtained through the use of less stringent criteria, called "fuzzy matching", using part-soundex code (first two characters of the four-character code) to overcome misspelling of surnames, and sex and date of birth. All data were anonymized to preserve the patients' identity, and their subsequent treatment locations were not identified.
For each LFU patient, we determined whether they had either subsequently registered and received care at another HIV clinic in the UK, and were therefore only LFU from King's College Hospital (Transfers), or had not registered at any other UK HIV clinic and were therefore lost to all clinical follow up in the UK (UKLFU). In order to ascertain whether UKLFU was due to unknown deaths (as reports on deaths for clinic attendees are based on passive reporting only), we also matched all LFU patients against the Office of National Statistics register of deaths in England, Wales and Northern Ireland for persons aged 60 years or younger, and through Health Protection Scotland using soundex code, sex and date of birth.
We matched our female UKLFU with the National Study of HIV in Pregnancy and Childhood using soundex code, date of birth, date of HIV diagnosis and country of birth in order to ascertain whether these patients may have been seen elsewhere for antenatal care subsequent to leaving the King's College Hospital HIV clinic. We also determined whether their HAART therapy was for the prevention of mother to child transmission alone. Finally, as a supplementary exercise, we reviewed the medical records for those defined as UKLFU to identify any documentation of intent to leave the UK, or other indications of reasons for LFU, and so further inform our proposed strategies for addressing UKLFU.

Statistical analyses
Demographic and clinical information, including gender, risk group, ethnicity, serial CD4 cell count and viral load, HAART history and dates of all clinic visits, were obtained from the HIV clinic electronic database for all patients. We compared the clinical characteristics of patients LFU from King's College Hospital HIV clinic (Transfers) and the UK (UKLFU), with all current clinic attendees (CCAs) at the King's College Hospital HIV clinic using Chi square or Fisher's exact tests for categorical variables, and Kruskal-Wallis rank test for continuous variables. Univariate and multivariate logistic regression analyses were performed to identify factors associated with UKLFU versus CCAs. Key variables included in the models were: year of clinic registration; age at HIV diagnosis; gender; risk group; ethnicity; use of HAART; initial clinical stage and CD4 count at clinic registration and also prior to LFU (or 31 December 2006 for those who were CCAs); and viral load at last visit prior to LFU for those on HAART.
We also repeated our analyses using a clinically relevant composite variable of gender-risk group and ethnicity, for descriptive ( Figure 1 and Table 1) and adjusted analyses (Tables 2 and 3) to help identify the sub-group at greatest risk of UKLFU and to inform strategies to reduce UKLFU. Since we hypothesized that factors associated with LFU would differ according to whether patients were receiving HAART, analyses were further stratified according to whether they had been prescribed HAART within six months of their last clinic appointment prior to LFU, or at 31 December 2006 for those who were CCAs. All statistical analyses were performed using STATA 10.0 software (STATA Corp., Texas, USA).

Results
There were 1859 HIV-positive patients aged 18 years or older who had registered and attended the King's College Hospital HIV clinic on at least one occasion over a nine-year period (between 1 January 1997 and 31 December 2005). Overall, there were 625 (33.6%) black African or Caribbean women and 348 (18.7%) black African or Caribbean men, and 445 (23.9%) white men who have sex with men (MSM). Less than 5% were either white heterosexual men or women. The median age at HIV diagnosis was 32 years (IQR 27-37).
In all, 432 (23.2%) patients were registered at the King's College Hospital HIV clinic between 1997 and 1999, 666 (35.8%) between 2000 and 2002, and 761 (40.9%) between 2003 and 2005. The initial median CD4 cell count was 303 (IQR: 141-467). In total, 762 (40.1%) had been prescribed HAART within six months of their last clinic visit. Those prescribed HAART within six months of their last clinic visit were more likely to be white than those who had not received HAART, even after adjustment for CD4 count, but there was no significant difference in gender or risk group.
A total of 722 (38.8%) of 1859 patients were defined as LFU as they had not been seen at the clinic for a year or longer (23.3% and 61.1%, respectively, for those receiving and not receiving HAART; Figure 2). Of these 722, 347 (48.1%), or 18.7% of all registered patients, were defined as Transfers (49.2% and 47.4% of those patients LFU receiving or not receiving HAART, respectively) as they had subsequently been seen at another HIV clinic in the UK, and 12 of these (3.5%) had died.
In total, 375 (51.9% of 722 LFU patients, or 20.2% of the 1859 eligible clinic population) were defined as UKLFU because there was no record of any further clinical follow up in the UK (50.8% and 52.6% of those patients LFU receiving or not receiving HAART, respectively). Overall, nine (2.4%) had died. After exclusion of 33 deceased patients (12 CCAs, 12 Transfers and nine UKLFU), subsequent analyses were based on the remaining 366 UKLFU patients (125 on HAART and 241 not on HAART) apparently lost to all follow up in the UK, and therefore at the greatest risk of subsequent HIV-related morbidity and mortality.
Frequency of LFU according to gender-ethnicity risk group Figure 1 shows the percentages of Transfers (n = 326) and UKLFU (n = 360) according to eight ethnicitygender HIV risk group categories. The highest percentage of Transfers to another HIV clinic in the UK was among white MSM (29.2%), followed by white women (24.1%), black African or Caribbean MSM (17.1%), and white heterosexual men (15.1%). In contrast, UKLFU was lowest among white and black MSM (11.2% and 14.3%, respectively), and white women (14.8%); it was highest among black Caribbean (29.3%) and African (26.2%) men, followed by black Caribbean and African women (23.4% and 22.1%, respectively).   Our overall rate of one in five becoming lost to UK follow up (after exclusion of deaths) was higher than the 8.5% [21] and 11.9% [23] from two cohorts in France, but was comparable to the 25% reported by an Italian cohort with a high proportion of injecting drug users [26], the 27% reported by a Boston clinical cohort [22],  [24]), and the 16% reported in a survey of community-based settings led by the American Foundation of AIDS Research [25]. These differences were not explained by significant variations in definitions of LFU, as the majority defined LFU as patients who had not been seen in a clinic for at least 12 months [22][23][24][26][27][28]. However, the lower rates of LFU reported in one of the French cohorts [21] may be due to the shorter, oneyear follow-up period compared with a cumulative LFU rate over a nine-year period in our study. Our overall high rate of loss to follow up also largely reflects the high proportion (more than 50%) of migrants originating from sub-Saharan Africa or the Caribbean in our clinic population, with high mobility and the highest rates of loss to follow up. The medical records review confirmed this, with documentation indicating that a high proportion planned to leave the UK, either voluntarily or because of immigration problems. In contrast, among white and black MSM, and white heterosexual women and men, the rates of UKLFU were less than 18%.

Characteristics of Transfers and UKLFU versus current clinic attendees
It is possible that we may have overestimated the proportion of HIV-infected patients LFU in the UK for several reasons. There may have been difficulties obtaining an exact match with the national SOPHID databases because patients may have registered at another site using a different date of birth or soundex code (either due to registration using different identifiers or due to coding differences, particularly in the transcribing and coding of uncommon surnames). In addition, the number of patients LFU who were deceased may have been underestimated since the national death register could only match deceased patients younger than 60 years of age. However, this would not have contributed significantly to a mismatch since only 10 of 366 (< 3%) UKLFU patients were older than 60 years.
Finally, we were unable to confirm whether patients UKLFU in the UK had in fact left the country and were either receiving ongoing care or had died in another country. In approximately half of those with available documentation, there was some indication that they either intended to leave the UK or that there were immigration problems, suggesting that they might have been deported. However, this information was not collected systematically, and was also available on only a subset of patients, and should therefore be interpreted with caution.
Therefore, among both HAART recipients and nonrecipients, it is the most vulnerable patients who are most likely to be lost to follow-up. Vulnerable patients encompass those with advanced immunodeficiency, at the highest risk of disease progression and in need of ongoing care, as well as those on HAART, with poorly controlled viraemia with the added risk for onward HIV transmission of potentially HIV drug-resistant virus.
Our findings also show that black African men are the most likely to default from care before receiving HAART, consistent with studies from sub-Saharan Africa that have highlighted the challenges of engaging African men with HIV testing and clinical services [32][33][34]. In contrast, among those receiving HAART, the highest probability of loss to follow up was among African women (rather than men). Of note, this was not explained by pregnant women receiving short-term HAART for prevention of mother to child transmission and then defaulting, as this accounted for less than 10% of these women. In addition, the median follow up for UKLFU was only 2.2 months for those not on HAART and 9.9 months for those recently prescribed HAART. This is similar to the findings from patient cohorts from France [23] and Italy [26], where the majority were LFU within six months of diagnosis. Nearly half of UKLFU patients not yet on HAART had made two or less clinic visits when they defaulted (usually following initial HIV testing and diagnosis). This highlights the importance of early supportive intervention in the weeks after initial diagnosis in prevention of early defaulting from care.
Several important measures to minimize LFU have been implemented at the clinic in the past 18 months. These include: text messaging reminders to patients' mobile phones the day before scheduled appointments; reducing the number of initial clinic visits with different members of the multi-disciplinary team following diagnosis; regular review of patients with missed appointments to allow early contact and intervention; and the appointment of a peer-support worker. However, formal prospective evaluation of the relative impact of these different strategies to reduce loss to follow up among HIV patients is now urgently needed in both developed and resource-limited settings.
A further important implication of our findings of high rates of loss to follow up is on the interpretation of outcomes, such as toxicity, rates of virological suppression, adherence and mortality, reported from clinical cohorts, which are based on those patients remaining in follow up, and may therefore substantially overestimate the proportion with a favourable outcome.

Conclusions
We observed a high rate of loss to follow up from our HIV clinic in south London, and this was highest among black African men and women. Our results highlight the need to better understand the health-seeking behaviours of patients LFU and to implement strategies in HIV clinics for both better tracking and minimizing of loss to follow up from HIV care.