Results of the search

The updated literature search to May 2020 identified 425 records: CENTRAL, 117 references, MEDLINE, 87 references; Embase, 221 references. Following preliminary deduplication across the databases, the combined total was 383 references.

We utilised the Cochrane author support tool Covidence for title and abstract screening of the 383 references.

Two review authors (DMF and MW) independently examined the remaining 20 potential studies. We excluded those studies that clearly did not meet the inclusion criteria, and obtained full‐text copies of five potentially relevant references. Subsequently, no additional studies were identified for inclusion (Figure 1; Figure 2).

Figure 1

---

---

Study flow diagram.

Figure 2

---

---

Study flow diagram.

We subsequently re reviewed all ongoing studies identified in the 2018 review (Jenkinson 2018) and updated with additional newly identified ongoing studies (NCT01811121 (RESECT), NCT02150564, NCT03291977 (FLEGME), NCT03531333). One author with an ongoing trial has published several interim analyses (NCT01479686), but the final results remain unpublished. Consequently, this was not included in this version of the review.

Included studies

The four included studies are Kubben 2014; Senft 2011; Stummer 2006 and Willems 2006 and for the purposes of consideration for a NMA are redescribed in detail below and in the Characteristics of included studies table.

In summary, we identified two trials of iMRI (Kubben 2014; Senft 2011), one trial of fluorescence‐guided surgery (Stummer 2006), and one trial of neuronavigation (Willems 2006). We found no eligible studies of ultrasound‐guided surgery.

Excluded studies

We excluded 32 studies, as follows (see Characteristics of excluded studies).

• Five studies were duplicates between searches.

• Nine studies were classified as ongoing (see Characteristics of ongoing studies).

• Twelve studies were not RCTs (Czyz 2011; Golub 2020; Koc 2008; Stepp 2007; Wu 2003; Wu 2004; Wu 2007; Zhang 2015; Abraham 2018; Abraham 2019; Wadhwa 2019; Waqas 2018).

• Three were only presented as abstracts and we were unable to obtain sufficient information even after attempting correspondence with the original trial authors (Chen 2011; Chen 2012; Seddighi 2016).

• Three did not directly compare an intraoperative imaging intervention with either another intraoperative imaging intervention or standard surgery (Eljamel 2008; Rohde 2011; Stummer 2017).

All ongoing studies were reviewed, with further written communication in relation to one ongoing RCT performed on 29 June 2020 (NCT01479686). Due to its current status as an active ongoing trial with planned publication of the final report at the time of writing this review, this study is listed as ongoing and will be considered when results are available.

Allocation

Blinding

Three trials performed blinded assessment for extent of resection (Kubben 2014; Senft 2011; Stummer 2006), and one trial for histological assessment (Stummer 2006). Regarding OS, blinding would not affect outcome reporting but could affect subsequent treatment. For QoL, PFS, and AEs, blinding would likely affect the outcomes reported. All trials were not blinded to participants or clinicians.

Incomplete outcome data

One trial accounted for all participants (Kubben 2014). Two trials accounted for all participants, but did not perform an intention‐to‐treat analysis, as those participants that had alternative pathological diagnoses were excluded (Senft 2011; Stummer 2006). In the remaining trial there was evidence of attrition bias for extent of resection (analysis of 32/42 participants) (Willems 2006).

Selective reporting

One trial reported all outcomes and was, therefore, at low risk of reporting bias (Senft 2011). Selective outcome reporting was apparent in three trials: one trial did not report QoL outcomes (Kubben 2014); one trial did not report full outcome data in the form of figures and appropriate statistics for survival, PFS, and AEs for 5‐ALA (Stummer 2006); and one trial did not present full data for survival, QoL, or AEs (Willems 2006). AE data in all studies were particularly poorly reported in terms of total number of events, number of participants with multiple events, and timing of events.

Other potential sources of bias

One of the issues with iMRI is attribution bias. Because surgeons know they can check for residual disease, they do not operate as aggressively as they might if they could not check for residual disease during the operation. So when a scan is done, residual disease is more likely to be detected, removed, and the success of the removal attributed to the iMRI. This is likely to affect outcomes that report a difference between the first intraoperative and final postoperative MRI scans.

Extent of resection

Extent of resection was reported as proportion with incomplete resection in each arm. The RR for the extent of resection in participants with glioma favoured the experimental arms in two of the four trials reporting this outcome, indicating a lower risk of having an incomplete resection with the intervention:

• iMRI: Senft 2011 achieved complete tumour resection in 23/24 (96%) participants in the intervention group compared with 17/25 (68%) participants in the control group (RR for incomplete resection 0.13, 95% CI 0.02 to 0.96; very low‐quality evidence). Kubben 2014 reported tumour resection using residual tumour volume and data for complete tumour resections were not available.

• 5‐ALA: Stummer 2006 performed complete resection in 90/139 (65%) participants in the intervention group versus 47/131 (36%) participants in the control group (RR for incomplete resection 0.55, 95% CI 0.42 to 0.71; low‐quality evidence).

• Neuronavigation: Willems 2006 achieved complete resection in three participants in the control group and five participants in the neuronavigation group. However, there was significant attrition, with not all participants having complete imaging, and the denominators for these figures were not stated, precluding meta‐analysis (very low‐quality evidence).

Adverse events

Reporting of AEs was inconsistent between trials and not according to the prespecified manner required in our protocol (Fountain 2020). Specifically, data were not available for participants at risk, participants with multiple events, timing of events, and outcomes of events. Therefore, we adopted a descriptive method using the data available to describe the AEs in each trial.

• iMRI: in the trial of Senft 2011, new or aggravated neurological deficits were present in 2/25 (8%) participants in the control group and 3/24 (13%) participants in the iMRI group; intraoperative imaging did not lead to continuation of tumour resection in any of the participants with AEs. Two participants had symptomatic haematomas, which were not attributable to the use of iMRI. In one participant, hemianopia was deliberately accepted due to tumour extension around the temporal horn of the lateral ventricle involving the optic radiation. In the Kubben 2014 trial, one participant in the intervention arm experienced a postoperative haemorrhage.

• 5‐ALA: AEs were present in 58.7% of the intervention arm versus 57.8% of the control arm in Stummer 2006. Neurological AEs were present in 42.8% of the intervention arm (7.0% grade 3 to 4) and 44.5% of the control arm (5.2% grade 3 to 4). There were significant neurological AEs in 12.4% of the intervention arm versus 11.6% of the control arm. The number of participants with a deterioration in the National Institutes of Health Stroke Scale compared to baseline tended to be higher in the intervention arm at 48 hours (26.2% with 5‐ALA versus 14.5% with control) but not at seven days (20.5% with 5‐ALA versus 10.7% with control), six weeks (17.1% with 5‐ALA versus 11.3% with control), and three months (19.6% with 5‐ALA versus 18.6% with control). No denominators were given for each result, preventing calculation of RRs and CIs.

• Neuronavigation: new or worsened neurological deficits were present at three months in 45.5% of participants in the control group and 18.2% of participants in the neuronavigation group in the Willems 2006 trial. During the first three months after surgery, seven participants (31.8%) in the control group and seven (30.4%) in the neuronavigation group experienced a new, non‐neurological AE. In three participants in the neuronavigation group, these events were fatal (pulmonary embolism, cardiac arrest with pulseless electrical activity, and postoperative pulmonary insufficiency). Other AEs included pulmonary or urinary tract infection, surgical removal of an epidural haematoma, surgical cyst drainage, repeated tumour debulking, CSF leakage, postoperative delirium, and insufficiently treated steroid‐induced diabetes.

Overall survival

• iMRI: Senft 2011 did not assess OS in the initial publication, although an abstract was published in 2017 included 24/29 (83%) of patients allocated to iMRI group and 21/29 (72.4%) of controls and reported that "iMRI itself did not affect outcome (560 versus 624 days, P = 0.53)". Kubben 2014 did not report OS in the prespecified manner for inclusion.

• 5‐ALA: in Stummer 2006, there was no difference in OS between the intervention and control arms (HR 0.82, 95% CI 0.62 to 1.07). Median survival was also reported and this was 15.2 months (95% CI 12.9 to 17.5) in the intervention arm versus 13.5 months (95% CI 12.0 to 14.7) in the control arm.

• Neuronavigation: Willems 2006 reported the HR to be 1.6, however, no CIs were available or able to be calculated. The median survival time was reported to be nine months in the control arm and 5.6 months in the intervention arm.

Progression‐free survival or time to progression

• iMRI: In Senft 2011, HRs or their respective CIs were not available and could not be calculated. The median PFS in the intervention arm was 226 days (95% CI 0.0 to 454) versus 154 days (95% CI 60 to 248) in the control arm. Kubben 2014 did not assess these outcomes.

• 5‐ALA: HRs and their respective CIs were not available and could not be calculated in Stummer 2006. Median PFS was 5.1 months (95% CI 3.4 to 6.0) in the intervention arm versus 3.6 months (95% CI 3.2 to 4.4 months) in the control arm.

• Neuronavigation: Willems 2006 did not assess time to progression or PFS.

Quality of life

• iMRI: Senft 2011 and Kubben 2014 did not report data for QoL.

• 5‐ALA: Stummer 2006 did not assess QoL.

• Neuronavigation: In Willems 2006, QoL questionnaires at three months' postoperatively were completed by 19 participants (eight in the neuronavigation arm and 11 in the standard surgery arm), constituting 64.5% of all eligible participants. The questionnaire included the EORTC QLQ‐C30 with 30 general questions, with additionally the 20‐question brain tumour module (BN20). Out of 26 outcome measures that were presented, the direction of change differed in seven (all in the BN20 group): four were in favour of the neuronavigation group and three were in favour of standard surgery. No statistical analysis was presented.

We considered this evidence to be of low to very low quality for all reported outcomes (summary of findings Table 1; summary of findings Table 2; summary of findings Table 3).

For all effects of interventions, meta‐analysis and NMA were not performed. Although all unit of analyses were appropriate, the included studies were excessively heterogeneous clinically for pair‐wise comparison, and there was an insufficient number of studies across interventions with low sample sizes with variations in the image guidance tools used in the control arms (predominantly utilisation of neuronavigation).

Brief economic commentary

To supplement the main systematic review of effects, we sought to identify cost analyses and economic evaluations that compared the interventions with each other or between different variants of the same intervention. A search of MEDLINE and Embase identified seven such studies (Abraham 2019; Eljamel 2016; Esteves 2015a; Hall 2003; Kowalik 2000; Makary 2011; Schulder 2003; Slof 2015) (one study was reported in two papers – Hall 2003; Kowalik 2000). Of the seven studies, four studies compared iMRI to conventional surgery (Abraham 2019; Kowalik 2000; Makary 2011; Schulder 2003); two compared 5‐ALA with white light surgery (Esteves 2015a; Slof 2015); and one compared conventional, 5‐ALA, fluorescein, ultrasound, and iMRI surgery (Eljamel 2016). Three studies were conducted in the USA (Kowalik 2000; Makary 2011; Schulder 2003), one in Portugal (Esteves 2015a), and one in Spain (Slof 2015), and for two it was unclear (but was probably the USA) (Abraham 2019; Eljamel 2016). Four studies were based on non‐randomised retrospective comparative cohorts (Kowalik 2000; Makary 2011; Schulder 2003; Slof 2015); one was based on a review and pair‐wise meta‐analyses (Eljamel 2016), and one used data from a trial and retrospective cohorts (Esteves 2015a). One study parameterised their microsimulation model with data from prospective and retrospective cohorts studies and a randomised trial (Abraham 2019). Costs estimates were derived from existing cost analysis and cost‐effectiveness analysis (including Eljamel 2016). Utility data were derived from a previous health technology assessment and preference elicitation studies, all utility data were derived using the standard gamble method. The studies based on single cohort studies all involved fewer than 100 participants, except for Slof 2015, which included 254 participants who received 5‐ALA and 120 who received white light surgery. All the studies except two (Abraham 2019; Esteves 2015a;), which integrated data using a Markov model and a microsimulation model respectively, were based on comparisons of individual patient level data. In terms of costs, what costs were included and over what time horizon varied markedly. Two studies considered costs over the patient lifetime (Abraham 2019; Esteves 2015a), and one only considered the drug cost (Slof 2015). The other studies considered costs incurred in hospital for the index surgery. Costs were reported in USD in five studies (Abraham 2019; Esteves 2015a; Kowalik 2000; Makary 2011; Schulder 2003), but the price year was stated only in two studies (Abraham 2019; Makary 2011). The other two studies reported costs in EUR, and the price year was 2012 in one study (Esteves 2015a), and not stated in the other (Slof 2015). Two studies were cost analyses only (Kowalik 2000; Schulder 2003). Effects were resection rates (Eljamel 2016), resection‐free years (Makary 2011), quality‐adjusted life years (QALY) (Abraham 2019; Eljamel 2016; Esteves 2015a; Slof 2015), PFS (Esteves 2015a), and life years (Esteves 2015a).

For the comparison of iMRI with conventional surgery, two studies reported a potential cost saving driven by reductions in length of stay (Kowalik 2000; Schulder 2003), and third study reported lower mean costs that were not statistically significant (Makary 2011). The one cost‐effectiveness analysis (Makary 2011) reported a lower complication rate for iMRI versus cMRI in people presenting for their initial tumour resection as well as a longer interval to repeat resection (20.1 months versus 6.7 months; P = 0.02); further results suggested iMRI was more cost‐effective in terms of cost per resection‐free years. Another study reported that iMRI was the most costly of conventional, 5‐ALA, fluorescein, and ultrasound‐assisted surgery (Eljamel 2016). iMRI was the least cost‐effective, but the results could not be replicated from the data presented in the study. Estimates of cost‐effectiveness (and cost over a longer follow‐up) need to considered due to the very limited evidence for iMRI where there is a benefit shown in terms of extent of resection but no evidence in the review of clinical effectiveness on OS. The one cost utility analysis was based on a microsimulation model and simulated cost and outcomes for a hypothetical cohort of 100,000 participants in each arm, the authors did not discuss the rationale for this sample size (Abraham 2019). The authors found that iMRI yielded an incremental benefit of 0.18 QALYs at an incremental cost of USD 13,447, which resulting in an incremental cost‐effectiveness ratio of USD 76,442 per QALY. The authors concluded that there was a 99.5% chance of cost‐effectiveness at a willingness‐to‐pay threshold of USD 100,000 per QALY based on probabilistic sensitivity analysis. The authors did not discuss the reasons for selecting a USD 100,000 per QALY cost‐effectiveness threshold.

For the comparison of 5‐ALA and standard surgery, 5‐ALA was on average more costly in both studies, but results in more quality adjusted life years over the patient lifetime or over the time to progression of disease (Esteves 2015a; Slof 2015). In both cases, the study authors concluded that the extra costs were worth the extra QALYs and that these conclusions were consistent over all sensitivity analyses conducted. These findings of extra effectiveness in the economic studies need to be considered in context of the findings of the review of the best available clinical effectiveness data summarised above.

We did not subject the identified cost analyses and economic evaluations to critical appraisal and we did not attempt to draw any firm or general conclusions regarding the relative costs or efficiency of the iMRI strategies compared. The evidence seems to state there is additional benefit in the additional intraoperative imaging strategies but the value of this benefit seems to vary. The evidence needs to be considered in the context of the decision makers' local context and with different national acceptability thresholds for cost‐effectiveness.