Ventriculoperitoneal shunt insertion in human immunodeficiency virus infected adults: a systematic review and meta-analysis

The following electronic online databases were systematically searched using terms for ventriculoperitoneal shunting or CSF diversion and HIV infection (see supplementary materials for full search strategies): The Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE (PubMed), EMBASE, CINAHL Plus (EBSCOhost), LILACS (BIREME), Research Registry (www.​researchregistry​.​com), the metaRegister of Controlled Trials (mRCT) (www.​controlled-trials.​com), ClinicalTrials.​gov (www.​clinicaltrials.​gov) and African Journals Online (AJOL). Grey Literature searching will be performed using the OpenGREY database. These were last updated on 17 October 2019. Our search sensitivity was validated confirming that candidate studies identified during scoping searches are included in the search yields [20, 44‐46].

The reference lists of included studies were scrutinized to identify any additional studies for inclusion. Two experts in relevant fields (GM and GF) were consulted to identify key studies not identified by electronic searches. No hand searching was employed.

Each abstract was independently screened against eligibility criteria by two of three reviewers (JL, NM, MP, ST). Where eligibility criteria were met, or possibly met, the full manuscript was obtained and reviewed by two reviewers to determine final inclusion. Any disagreements were resolved by consensus discussion.

Data was extracted using standardised proformas (supplementary materials) and entered into Review Manager [47]. Reference management was done using EndNote [48].

Each included study was initially assessed independently for risk of bias by two reviewers (JL, MP, ST) using the Canadian National Collaborating Centre for Methods and Tools domain based “Quality Assessment Tool for Quantitative Studies” [49, 50]. Disagreement was resolved by consensus discussion.

Our narrative review summarises each included study’s reported measures of effect [39]. For quantitative meta-analysis, we report pooled odds ratios [51].

The corresponding authors of all published abstracts of studies that were potentially eligible for inclusion, but for which a full study report was not published, were contacted electronically to request data not provided by the abstract. All authors contacted were allowed a minimum of 18 months to respond, prior to submission of our completed article.

On the basis of extracted data, included studies were compared for clinical and methodological heterogeneity. Clinically and methodologically studies reporting similar outcome measures at similar time points, were assessed for statistical heterogeneity by calculation of the I² statistic (%) [47, 51]. We defined substantial heterogeneity as I² greater than 50%.

Of the excluded studies, 16 concerned CM [54‐57, 59‐63, 66‐69, 76, 78, 80], four TBM [20, 58, 64, 77], two bacterial meningitis [70, 71], one coccoidal meningitis [8], and one included only cases of diagnostic uncertainty [65].

Two studies were published only in abstract form and did not present sufficient outcome data for inclusion [67, 68]. The corresponding authors of these records were therefore contacted to request further details to assess eligibility [8, 67, 68]. The authors one of these responded but were unable to provide further information [8, 68]. This study was therefore excluded.

The reasons for exclusion from this review are as follows: Two studies included patients with multiple, undocumented modalities of CSF diversion [54, 55]; one study provided a just single case report [56]; three studies reported just one or two cases of surgical intervention and no survival/outcome data [57, 60, 61]; one study included 44 patients who underwent VPS in the context of CM but no outcome data [80]; two studies were review articles [58, 66]; eight reported no HIV-infected patients who underwent VPS [20, 59, 62‐65, 71, 76‐78]; in one study reporting < 50% HIV-infected patients it was not possible to associate outcome with HIV status [8] and; no relevant measures of survival/outcome were studied in four [67‐70].

This was a retrospective consecutive interrupted time series of all patients admitted with severe intracranial hypertension (lumbar puncture opening pressure > 35cmCSF) secondary to CM who received CSF diversion. Four patients received a VP shunt. Three patients were initially managed with serial lumbar puncture. One received primary VPS. All patients presented with visual disturbance, one had seizures and one had bilateral abducens nerve palsies. Initial Glasgow Coma Score (GCS) was not reported. Neither makes nor models of the VPS systems used were described.

Although this study included consecutive patients and therefore minimised selection bias, it was classified as being globally weak. An interrupted time series is a weak study design. Coupled with small sample size, this rendered meaningful multivariable analysis impossible and it was not attempted. The time until loss to follow up of one patient was unclear, and the reasons for this are not documented. Mortality is a robust outcome measure.

This is an interrupted time series that presents five HIV-infected patients who were treated with VPS whilst admitted to a single Intensive Care Unit (ICU) in Montevideo, Uruguay. It is unclear if the data was collected in a prospective or retrospective fashion or whether patients were recruited consecutively. Median initial GCS was 12 (range 9–15). CD4+ cell count was not reported for any patient, nor was the make or model of VPS system. The VPS was inserted as a primary procedure for four patients with CSF pressures >23cmH₂O. One patient initially received an EVD with intraventricular pressure monitoring before having a VPS inserted. All patients were managed with antifungal agents and hyperventilation to a target pCO₂ of 30 mmHg. Mannitol boluses were used in an unclear number of patients. At discharge from ICU all five patients were alive. Time to discharge from ICU was not reported. At 6 months two patients had been lost to follow-up and three survived. Outcome in this study is described as “good recovery”, “moderately disabled” or “dead” but the criteria for assignment to these descriptors are not described. At ICU discharge, 4 patients had made a “good recovery” and one was “moderately disabled”. At 3 months, two had a “good recovery” and one was “moderately disabled”. No complications or shunt failure are reported.

This study was classified as being globally weak with a high risk of selection bias and weak study design which, coupled with small sample size, did not permit adjustment for modulators of risk of mortality or poor outcome. Outcome assessment did not use a validated outcome measure. 40% of patients were lost to follow-up at 6 months.

This two-cohort analysis reported all patients with CM and CM IRIS presenting to a single hospital in Texas, USA. 49 of 50 patients had associated HIV infection. Fourty patients had communicating hydrocephalus, and it is unclear if the patient with non-communicating hydrocephalus was HIV-infected or not. CD4+ cell count was not documented for any patient, nor was initial neurological status. All further demographic and outcome data are specific for the HIV-infected group. Patients who went on to receive CSF diversion had initial lumbar puncture opening pressure > 25 cm H₂O. They were initially managed with antifungals, antiretroviral therapy and a variable number of therapeutic lumbar punctures (median 7; range 2–40) for a variable period (median 27 days; range 4–281 days) before proceeding to VPS. Eight patients underwent VPS.

At 1 month, five of the eight VPS patients survived. Three deaths had occurred, one due to post-extubation aspiration pneumonia, one due to septic shock of undocumented aetiology and the other due to a gram-negative bacterial pneumonia. At 6 months, five patients were still alive, and no new deaths were recorded. At 12 months, three patients were alive, four were dead and one was lost to follow-up. The additional death was secondary to an unspecified respiratory complication. Of the three patients initially treated for CM IRIS, at both one and 6 months, two were alive and one dead. At 12 months, two were confirmed dead and the other lost to follow-up.

One instance of shunt infection was recorded at 12 months. Two other patients, who were followed up for 1512 and 1485 days, suffered a shunt occlusion and a shunt infection, respectively, at some point during their follow-up period. However, the time to these complications was not reported. One other patient who was followed up for 383 days suffered a subdural haematoma, which was attributed to VPS over-drainage. Again, the time point at which this occurred was not reported. Functional outcomes were not reported.

This study was classified as being globally weak with no attempt to undertake multivariable survival analysis, undocumented reasons for loss to follow-up and lack of reporting of times of complications. Survival is a validated outcome measure but was unreliably reported in patients discharged to hospice care. Rates of complication were not measured using a validated method. Patients were consecutively recruited, and the selection process was clearly documented. A two-cohort analysis study has moderate risk of bias.

This retrospective interrupted time series included 15 patients with CM who presented to medical services in Buenos Aires, Argentina. 14 of these patients were HIV-infected, however it was not possible to identify which demographic or outcome data pertained to the HIV-non-infected patient. CSF diversion was undertaken in patients who had lumbar puncture opening pressure persistently > 25 cm H₂0 after 2 weeks of therapeutic lumbar puncture and antifungal treatment. 14 of the 15 patients were managed with VP shunt, and one with lumboperitoneal shunting. Again, it was not possible to isolate demographic and outcome data from this patient. Medtronic® systems were used for all patients but component models are not described.

Three patients suffered complications including sepsis, multiorgan failure, two cases of meningitis and abdominal pseudocyst formation related to the distal catheter tip and requiring catheter revision. These occurred at unclear time points and the patients were excluded from survival analysis by the study authors. However, the authors did report that at least one of the patients with meningitis and multiorgan failure died. Of the remaining patients, all survived to 12 months. No other complications or outcomes were reported in this group.

This study was classified as being globally weak with high risk of selection bias, a weak study design and small sample size which did not permit adjustment for factors potentially confounding outcome, and poorly documented exclusion from follow up causing high risk of attrition bias. Survival is a validated outcome measure but was unreliably reported in the group with complications.

This retrospective single cohort study included nine patients with HIV-associated CM managed with VPS over a five-year period at a Shanghai hospital, China. Although exclusion criteria are documented, the number of exclusions is not. Included patients had a mean CD4+ cell count of 11 cells/μL. Neurosurgical intervention was considered in patients with raised lumbar puncture opening pressure and deteriorating neurological status despite antifungal therapy, repeated therapeutic lumbar puncture and mannitol administration. VPS was undertaken in all patients using a Medtronic® system, but the valve and catheter model are not documented. The authors define raised opening pressure as “> 400 cm H₂0” which, presumably, is a typographic error (raised CSF opening pressure defined as > 20-35 cm H₂O in other included studies [45, 53, 73‐75]). Two patients were described as having loss of consciousness at baseline, but no objective measures were provided.

Eight of the nine patients studied were alive at 1 month. Three patients were lost to follow up at 6 months and no new deaths were reported. At 12 months, one more patient had died, yielding a survival of four out of six patients for whom follow-up was available. Given the low initial CD4 counts, it is likely that deaths would have occurred in those lost to follow-up by this time point and therefore there is a risk of attrition bias at this time point. The causes of death were VPS obstruction at 1 month and an unspecified reaction to antiretroviral drugs at 12 months. No other outcome measures, shunt failures or complications were reported.

This study was classified as being globally weak due to a weak study design and unknown risk of selection bias. No confounders were identified, and the primary outcome was survival – which is a robust outcome measure.

This retrospective cohort study investigated associations between various parameters derived from quantitative CSF microscopy with mortality in a cohort of patients who presented to hospital in São Paulo, Brazil, with HIV-associated CM. Included patients were admitted between January 2006 and June 2008, but it is unclear what rate of case acquisition this represents. A median CD4+ cell count of 36 cells/μL, with interquartile range of 17–87 cells/μL was reported across all patients. Neurosurgical intervention was considered in patients with lumbar puncture opening pressure persistently >20cmH₂O after 14 days therapeutic lumbar puncture and antifungal treatment.

Although survival was presented in the series of 9 patients undergoing VPS, no other dependent or independent variables were reported in this subgroup to allow comparisons and therefore, for our purposes, this study provided case-series data. Initial neurological status and type of shunt valve or catheter were not described. Of the 9 patients who underwent VPS, 6 survived until hospital discharge. However, time to discharge and length of stay were not reported. No other outcomes were reported.

This study was classified as being globally weak with an unknown risk of selection bias and weak study design and size which did not permit adjustment for confounding variables regards our research question. Survival is a strong primary outcome measure.

This prospective cohort analysis compared outcome and mortality at 1 month following VPS between HIV-infected and HIV non-infected patients. Patients who underwent VPS for TBM, over an undefined recruitment period prior to commencement of the national South African antiretroviral treatment (ART) programme in 2004, were recruited from a single hospital in KwaZulu-Natal, South Africa. Of the 30 patients recruited, 15 were HIV-infected and 15 HIV non-infected. The two cohorts differed widely in age range with the HIV-infected group being mostly adult and the HIV non-infected group being mostly paediatric. Patients were diagnosed with hydrocephalus on the basis of CT, transcranial doppler or clinical observation in keeping with hydrocephalus and intraoperative ventricular pressure > 20cmCSF. Median initial GCS was 14 (range 9–15) in the HIV-infected group and 12 (range 9–15) in the HIV non-infected group. Use of EVD or therapeutic lumbar puncture was not documented for any patient. VPS was undertaken using a 102 cm catheter and low-profile, medium pressure Codman / Unishunt valve (Codman / Johnson & Johnson®, Raynham, MA).

At 1 month, 5/15 and 11/15 HIV-infected, and HIV non-infected patients were alive, respectively. 11/15 and 6/15 patients in the HIV-infected and HIV non-infected cohorts had an unfavourable Glasgow Outcome Score [43] (GOS 1–3) dichotomised GOS at 1 month. Multivariable analysis was not attempted. Univariate analysis revealed a significant association of CD4+ cell count and outcome (p = 0.031), but the magnitude of this association was not reported.

This study was classified as being globally weak. A prospective cohort study provides a moderate study design. The study included representative patients, but the control group was poorly age matched and this was not adjusted for. It is not clear that patients were enrolled consecutively, yielding an unknown risk of selection bias. The authors were not blinded to HIV status. Mortality is a robust primary outcome measure. As this study was undertaken prior to ART being made widely available in South Africa, its findings may not be generalisable to settings where ART is available [81].

This retrospective case-control study compared outcome and mortality between HIV-infected and HIV non-infected patients following VPS. All 30 eligible HIV-infected patients who underwent VPS at a single hospital in Bangalore, India, between June 2002 and October 2012 were included. These patients were compared against 30 HIV non-infected control patients that were matched for age, sex and clinical grade of TBM. Patients were included who “had symptoms of raised intracranial pressure, with clinical, radiological and CSF findings characteristic of TBM”. However, specific inclusion criteria were not documented. Initial GCS was dichotomized as 3–8 (unfavourable) and this was present in 13.3 and 3.3% in the HIV-infected and non-infected patients, respectively. The rest had GCS 9–15, which was classified as favourable. 23 and 22 patients in the HIV-infected and non-infected cohorts, respectively, had communicating hydrocephalus. Patients underwent VPS as a primary procedure unless extensive basal infarcts were evident on computed tomography (CT) scan of the brain. In these patients, a frontal EVD was placed and converted to VPS if improvement in “sensorium” was noted. The proportion of those who underwent primary VPS is not reported. VPS was with a medium-pressure Chhabra shunt (Surgiwear® Inc.). 9 HIV-infected patients and 4 HIV non-infected patients were lost to follow-up.

At 1 month, 13/21 HIV-infected and 20/26 HIV non-infected patients were alive. At 6 months, 9 HIV-infected and 18 HIV non-infected patients were alive. In the HIV-infected cohort, unfavourable initial GCS was associated with higher rates of mortality. GOS at 3 months was also dichotomized as unfavourable (1–3) or favourable (4, 5). 16/21 HIV-infected, and 9/26 HIV-non-infected patients had an unfavourable outcome at 3 months. Binomial logistic regression analysis confirmed that HIV infection (p = 0.038) and low Palur grade (p = 0.024) [82] were significantly associated with unfavourable outcome. No measure of effect was presented for these analyses. Anaemia (haemoglobin < 10 mg/dL) was associated with unfavourable outcome in the HIV-infected group (Exp.[β] = 25.6; p = 0.011).

This study was classified as being globally of moderate quality. A case-control study provides a moderate quality of study design. The study included representative consecutive cases with appropriate matching of controls. Potential confounders were adjusted for in the multivariable analysis. The authors were not blinded to HIV status. Just under a third of initially recruited patients were lost to follow-up at later time points, but this was appropriately documented. GCS and dichotomised GOS are robust outcome measures but not validated in the context of TBM. Mortality is a robust outcome measure.

This case control study compared 15 HIV-infected patients who were using ART prior to presentation who consecutively underwent VPS for TBM in 2017 with 15 historical retrospectively identified controls who underwent the same procedure but were not using ART. All patients receiving ART were prescribed single pill formulation of tenofovir and emtricitabine as well as efavirenz. The diagnosis of TBM was made on the basis of typical clinical, radiological and CSF findings. Not all patients had positive CSF cultures. All patients had radiological features of hydrocephalus with CSF pressures higher than 20 cm H₂O on ventriculostomy and underwent VPS as a primary procedure. Both cohorts had similar initial GCS and Palur grade on admission.

All patients were followed up as inpatient for a minimum of 1 month post-operatively. Outcome was assessed at discharge using the dichotomised GOS (GOS 4–5 good outcome; 1–3 poor outcome). At discharge overall 15 patients achieved good outcome, one poor outcome and four died. Eleven patients receiving antiretroviral therapy achieved a good outcome with no instances of poor outcome and four deaths. This was better than the historical control group only four of which achieved a good outcome, one poor outcome and ten deaths. This difference was statistically significant in unadjusted analysis (p = 0.027). Patients in the ART treated group were more likely to have a higher GCS at 1 month than at presentation than the ART untreated group (p = 0.042).

As two comparative cohort studies were included assessing the effect of HIV infection on outcome, it was possible to pool reported measures of effect of HIV-infection on survival at one-month post-VPS. Risk of bias is summarised in Table 3.

Nine studies of survival and/or outcome in TBM and CM were identified [44‐46, 53, 72‐75, 79]. However, none of these compared survival or outcome between TBM and CM. Study populations, outcome measures and follow-up varied greatly. Further, all of the studies of CM were of weak design, and two of the three studies of TBM also being graded as weak. There was significant methodological heterogeneity between CM and TBM studies and it was therefore not possible to conduct meaningful meta-analytic comparison of outcomes between these groups. We therefore present a narrative analysis. The results of pooling of reported outcomes within each group are presented in SOF tables, with the purpose of summarising reported outcomes to date. These are of very low to low GRADE of evidence (see risk of bias, above), indicating that the true outcomes for each pathology may be, or are likely to be, substantially different from the pooled measure [83]..

Three studies reported survival in CM at one-month post VPS [45, 72, 74]. Cherian, et al. reported three deaths in eight patients. Three of these patients had a diagnosis of CM IRIS, one of which died (AIDS-specific mortality). Liu, et al. report one death in nine patients at 1 month. This was due to shunt obstruction. Bach, et al.’s four patients with documented follow up were all alive at 1 month. In total, 17 of 21 (81%) patients described in these studies survived 1 month, with one episode of shunt failure documented.

At 6 months post-VPS, four studies reported survival [45, 72, 74, 75]. Three of Liu, et al’s patients had been lost to follow-up. However, no new deaths were confirmed. Bach, et al. reported one death due to HIV-wasting syndrome. Another had died of cytomegalovirus pancreatitis. These both therefore contributed to AIDS-specific mortality [84]. Calvo, et al. obtained 3 months follow up for three patients, all of whom survived. Cherian et al. followed up all of their eight patients and reported no new deaths. With loss to follow-up and addition of three surviving patients from Calvo, et al’s cohort, overall survival remained stable: 17 of 21 (81%) patients that were followed up at 6 months were alive. No new cases of shunt failure were documented.

Four studies reported 12-month survival [45, 72‐74]. Liu, et al. reported a further death. Cherian, et al. reported a death in one of their patients who had been diagnosed with CM IRIS, contributing to AIDS-specific mortality. Another of their patients was lost to follow-up. One of Bach et al.’s patients had died of Pneumocycstis jiroveci pneumonia, also contributing to AIDS-specific mortality. Corti, et al. report that all of their 12 patients followed-up for 1 year were alive. In total 20 of 29 patients with 12-month follow-up survived. Five of the 17 (29%) patients reported by studies describing cause of death had AIDS-defining illness as a primary or major contributing cause of death by 12 months.

Liu et al., reported one shunt failure at 1 month. Cherian and colleagues reported one shunt failure at 12 months, but as they also reported two other instances of shunt failure at unclear timepoints, their shunt failure data was not included in the pooled analysis.

Complications of treatment were reported by two studies [72, 74]. In Liu, et al.’s series, none of these were related to the surgical management of their patients at 12 months. In addition to their three instances of shunt failure, Cherian and colleagues reported one instance of over drainage subdural haematoma at an unspecified time. Follow-up in this series ranged from 28 to 1512 days.

No studies reported any measures of functional outcome at any time point. Calvo, et al. reported that two of three patients made a “good recovery” at 6 months but this measure is not validated [75].

Three studies reported 1-month survival following VPS for HIV-associated TBM [44, 46, 79]. Sharma et al. reported 13 of 21 patients surviving for one-month post-VPS. Nadvi, et al. reported just 5 of 15 patients surviving for one-month. Harrichandparsad, et al. reported 14 out of 30 patients surviving for 1 month, although 10 of these deaths were in the ART untreated group.

Sharma et al. reported 9 of 21 patients surviving 6 months. Sharma et al. also reported “long-term” survival of 7 of 21 with mean follow-up of 130 days, range 91–760 days. No studies reported causes of death or measures of shunt survival.

No studies reported rates of operative complication.

Two studies reported GOS in HIV-infected patients 1 month following VPS [46, 79]. 11 of 15 patients in Nadvi, et al.’s study had an unfavourable functional outcome, with 10 of these being dead. Therefore four of the five surviving patients attained GOS 4–5, signifying moderate to low levels of disability at one month [43]. Harrichandparsad, et al. reported 15 of 30 patients achieving an unfavourable outcome, including death at one month, 11 of which were in the ART untreated group [79].

The included studies of CM provided a very low level of evidence on which to inform clinical practice. Reported survival by included studies of CM varied extremely widely – from 33 to 100% at 1 year – and this is likely a consequence of study heterogeneity. The relatively high pooled 12-month survival of 69% might represent selection and attrition biases, which many included studies were considered to be at a high risk of. For comparison, 12-month survival in a study of 549 patients with HIV-associated CM who did not receive shunting varies from 55 to 93%, contingent on patients’ antiretroviral status [85].

The 2018 World Health Organisation (WHO) guidelines on CM recommend that CM-associated hydrocephalus be initially managed by therapeutic lumbar puncture and drainage of CSF [86]. These guidelines emphasise that repeat lumbar puncture should be undertaken to control symptomatic hydrocephalus until resolution and do not describe any role for permanent CSF diversion in CM. The Infectious Diseases Society of America 2010 guidelines for the management of cryptococcal disease recommended early ventricular CSF diversion for non-communicating hydrocephalus in CM [87]. For chronic communicating hydrocephalus or communicating hydrocephalus causing significant neurological impairment, permanent CSF diversion may be undertaken [87].

Three cohort studies were included. These provide a low level of evidence to inform prognostication at one-month, in terms of survival and functional outcome, for patients undergoing VPS for HIV-associated TBM. One-month survival varied from 33 to 61.9% between these studies, yielding a very cautious pooled estimate of one-month survival of 51.5% and a pooled estimate of 42.2% good outcome at 1 month. We anticipate that this estimate differs substantially from the true outcome. The included studies provide a very low level of evidence to inform later prognostication in terms of survival and functional outcome. For all other measures of outcome – including AIDS-specific mortality, shunt failure risk and perioperative complication – the included studies provided no data to inform clinical decision-making. Nadvi, et al. and Sharma, et al’s studies demonstrate that patients who underwent VPS for HIV-associated TBM had poorer outcomes than their HIV non-infected counterparts. Meta-analysis of these findings yielded increased risk of mortality (odds ratio 2.93) in the HIV-infected group. However, one of these studies compared a largely paediatric HIV non-infected population with a largely adult HIV-infected population [46]. The clinical phenotypes and pathophysiological responses to brain injury in these groups differ widely and so comparisons should be interpreted with caution [88, 89]. It is also important to note that whilst one of the studies employed ventricular CSF diversion as a primary measure for management of hydrocephalus [44], it is unclear if this was undertaken primarily, or following failure of lumbar drainage/medical management in the other study [46]. This heterogeneity limits the interpretation of comparisons between reported outcomes in HIV-infected and non-infected cohorts.

The British Infection Society published guidelines in 2009 for the management of TBM [90]. These guidelines did not recommend altering the medical management of patients with TBM on the basis of their HIV status but did not comment on their surgical management with respect to HIV. On the basis of observational studies of HIV non-infected patients, early VPS is recommended by these guidelines for non-communicating hydrocephalus [90‐92]. It is noted that response to external ventricular drainage poorly predicts response to VPS [91]. Our study’s findings would indicate that HIV status might impact significantly on surgical outcome and therefore caution should be employed by future guideline working groups before generalizing the results of studies conducted of HIV non-infected populations to inform the surgical management of HIV-infected patients.