Appendix MCost effectiveness of quantifying proteinuria in women with gestational hypertension

Publication Details

Introduction

The GDG initially compared the cost effectiveness of automated reagent-strip reading devices (automated urinalysis) with visual reading of reagent strips (dipsticks; visual urinalysis). The analysis presented in Appendix L suggested that automated urinalysis was more cost effective than visual urinalysis, and that formed the basis of the initial guideline recommendations. Following the pre-publication check, there were suggestions that protein : creatinine ratio (PCR) should be included as a comparator. It was thus agreed to undertake an additional analysis in which the the following screening methods for proteinuria in women with mild or moderate gestational hypertension were compared:

  • use of protein : creatinine ratio alone (PCR strategy)
  • use of an automated reagent-strip reading device followed by protein : creatinine ratio in women with a positive test result on the automated reagent-strip reading device (Auto + PCR strategy)
  • use of an automated reagent-strip reading device followed by a validated 24-hour urine collection in women with a positive test result on the automated reagent-strip reading device (Auto + 24-hour).

An automated reagent-strip reading device provides a point-of-care screening test, and a further gold standard test is needed to confirm the diagnosis. Traditionally, 24-hour urine collection has been regarded as the gold standard, but it has been suggested that PCR could fulfil this function. Thus, in the second and third strategies, PCR and 24-hour urine collection are considered to be the gold standard tests for quantifying proteinuria, respectively. However, PCR results can be available within a few hours and so the GDG also considered that PCR could be used directly in place of an initial screening test, in which case there would be no requirement for a confirmatory test.

The use of spot urinary PCR and spot urinary albumin : creatinine ratio (ACR) to estimate proteinuria is well established in the management of chronic kidney disease. More recently, it has started to be used in the management of hypertensive disorders during pregnancy, as in the case of the Australian and New Zealand guidelines on hypertension in pregnancy.254 The GDG’s view is that some tertiary centres in the UK use automated reagent-strip reading devices and PCR to screen for and quantify proteinuria.

Studies have shown that use of an automated reagent-strip reading device and PCR can improve the predictive power of urinalysis and eliminate the inter- and intra-observer variability that is present when visual dipstick urinalysis is used.83 Leanos-Miranda et al.91 suggested that PCR may be used as an alternative to 24-hour urine collection. However, the cost effectiveness of PCR alone, an automated reagent-strip reading device followed by PCR in women with a positive automated reagent-strip test result, or an automated reagent-strip reading device followed by 24-hour urine collection in women with a positive automated reagent-strip test result has not been evaluated. Practice varies within the NHS and the GDG estimates that approximately 20% of day assessment units use an automated reagent-strip reading device (based on a survey conducted by Action on Pre-Eclampsia) and that PCR is used in many centres (GDG opinion; PCR use was not evaluated in the Action on Pre-Eclampsia survey).

Aim

To determine the cost effectiveness of PCR alone, of an automated reagent-strip reading device followed by PCR in women with a positive automated reading device test result, and of an automated reagent-strip reading device followed by 24-hour urine collection in women with a positive automated reading device test result in screening for significant proteinuria in pregnant women with new-onset mild or moderate gestational hypertension.

Methods

Test parameters

The test parameters are shown in Table M.1. For the sensitivity and specificity of the automated reagent-strip reading device, we used data from the systematic review by Waugh et al.81 Data for PCR were taken from the five studies that assessed the accuracy of spot PCR compared with 24-hour urine collection for the screening and quantification of significant proteinuria in hypertensive pregnant women.90–94 The studies used different cut-off points and the five studies could not be meta-analysed owing to significant heterogeneity. Therefore, the results for each study were analysed separately.

Table M.1. Test accuracy statistics used in the health economic model for women with mild or moderate gestational hypertension.

Table M.1

Test accuracy statistics used in the health economic model for women with mild or moderate gestational hypertension.

Prevalence of pre-eclampsia

In the base-case analysis, a prevalence of 18% was assumed and various ranges were tested as part of sensitivity analysis (see Appendix K).

Clinical management

The clinical management of women in the model is described in Appendix K.

Cost parameters

The cost inputs used in the model are shown in Table M.2. It was assumed that any PCR false positives would be managed in the same way as true positives.

Table M.2. Health service costs incurred by women who have pre-eclampsia with mild or moderate hypertension, 2008–09.

Table M.2

Health service costs incurred by women who have pre-eclampsia with mild or moderate hypertension, 2008–09.

Estimation of QALY loss for false negatives

The estimation of QALY loss for false negatives is described in Appendix K.

Sensitivity analysis

Two-way sensitivity analysis was undertaken to assess the extent to which the results were affected by different test accuracy values.

Results

Tables M.3 and M.4 give the diagnostic outcomes and costs, respectively, of the diagnostic strategies using a ‘best-case’ scenario for PCR. Equivalent data for a ‘worst-case’ scenario for PCR are presented in Tables M.5 and M.6.

Table M.3. Diagnostic outcomes of PCR versus an automated reagent-strip reading device for 60 000 women with mild or moderate gestational hypertension using Leanos-Miranda et al.

Table M.3

Diagnostic outcomes of PCR versus an automated reagent-strip reading device for 60 000 women with mild or moderate gestational hypertension using Leanos-Miranda et al.

Table M.4. Screening strategy costs for 60 000 pregnant women with mild or moderate hypertension using Leanos-Miranda et al.

Table M.4

Screening strategy costs for 60 000 pregnant women with mild or moderate hypertension using Leanos-Miranda et al.

Table M.5. Diagnostic outcomes of PCR versus an automated reagent-strip reading device for 60 000 pregnant women with mild or moderate gestational hypertension using Dwyer et al.

Table M.5

Diagnostic outcomes of PCR versus an automated reagent-strip reading device for 60 000 pregnant women with mild or moderate gestational hypertension using Dwyer et al.

Table M.6. Screening strategy costs for 60 000 pregnant women with mild or moderate hypertension using Dwyer et al.

Table M.6

Screening strategy costs for 60 000 pregnant women with mild or moderate hypertension using Dwyer et al.

Summary cost-effectiveness results for all five studies are shown in Table M.7. The results suggest that PCR alone is the most cost-effective strategy using diagnostic accuracy data from Leonos-Miranda et al.91 for both moderate and mild gestational hypertension. In this case, PCR alone is said to dominate the other strategies because it is both less costly and more effective (generating the highest QALY gain). It should be noted that the cost and QALY gain were calculated relative to no screening, where all cases of disease are modelled as false negatives. The other four analyses, based on smaller studies, indicated that using the automated reagent-strip reading device followed by 24-hour urine collection would be cost effective for women with mild or moderate hypertension.

Table M.7. Cost effectiveness of screening strategies in 60 000 pregnant women with mild or moderate gestational hypertension.

Table M.7

Cost effectiveness of screening strategies in 60 000 pregnant women with mild or moderate gestational hypertension.

Sensitivity analysis

As the results in Table M.7 show, the relative cost effectiveness of PCR alone versus using an automated reagent-strip reading device followed by a confirmatory 24-hour urine collection is highly dependent on the sensitivity and specificity of the tests. To further analyse the impact of test uncertainty, a two-way sensitivity analysis was undertaken in which the sensitivity and specificity of PCR were varied between all possible pairwise combinations between 50% and 100% while holding all other model parameters constant, including the sensitivity and specificity of the automated reagent-strip reading device, at their baseline values. The two-way sensitivity analysis was restricted to PCR alone because using the automated reagent-strip reading device followed by PCR was dominated in both the best- and worst-case scenarios for PCR.

Figure M.1 shows how cost effectiveness varied across different PCR test characteristics. The results presented are for 60 000 pregnant women with moderate gestational hypertension and the results for mild disease would be similar.

two-way analysis

Figure M.1

Varying the sensitivity and specificity of PCR.

Figure M.1 is divided into four quadrants (A–D). The x-axis and y-axis represent sensitivity and specificity, respectively. The vertical and horizontal black lines represent the sensitivity (82%) and specificity (81%) of the automated reagent-strip reading device.

In quadrant A, the sensitivity of PCR is always less than or equal to the sensitivity of automated urinalysis. Also, the specificity of PCR is always more than or equal to the specificity of the automated reagent-strip reading device. This quadrant shows that, unless the specificity of PCR is considerably better than that of the automated reagent-strip reading device, PCR alone is dominated. This is because lower sensitivity means that there are more false negatives resulting in a lower QALY gain and also because of high treatment costs associated with a greater number of false positives and false negatives. Although in this quadrant the automated reagent-strip reading device has a higher false positive rate, these are identified by the confirmatory 24-hour urine collection and this limits unnecessary treatment. As the specificity of PCR rises, the cost of false positives falls until a point is reached when PCR alone becomes the cheapest strategy. However, even then, automated urinalysis would be preferred on cost-effectiveness grounds unless the sensitivity of PCR approaches that of the automated reagent-strip reading device.

Quadrant B represents scenarios where the test characteristics of PCR have better sensitivity and better specificity. In this quadrant, PCR alone will always produce the greater QALY gain as this is driven by test sensitivity. As in quadrant A, PCR may also have cost advantages at high specificities and this explains its dominant portion of this quadrant. As specificity falls, a point is reached where PCR alone becomes the more expensive strategy and in that case cost effectiveness is determined by whether the incremental QALY gain from PCR alone can be delivered at an acceptable incremental cost (i.e. at under £20,000 per QALY).

In Quadrant C, the lower specificity of PCR means that the incremental benefits arising from higher PCR sensitivity can never be justified by the incremental costs.

In Quadrant D, automated urinalysis has unambiguously better test characteristics with greater or equal sensitivity and specificity. In this quadrant, PCR is dominated because it has a lower QALY gain as a result of a lower sensitivity and a high cost of false positives and false negatives.

Discussion

The estimated sensitivities and specificities for PCR were obtained from five different studies that were not meta-analysed owing to heterogeneity. However, running the analysis for these studies separately showed that the cost-effectiveness results were sensitive to the accuracy of the respective tests. Where the most favourable PCR test accuracy data were used, PCR alone was shown to be the most cost-effective strategy, and this analysis was based on the largest of the five studies. However, when PCR sensitivity and specificity were derived from other studies, the use of an automated reagent-strip reading device followed by 24-hour urine collection was shown to be cost effective.

The use of an automated reagent-strip reading device followed by PCR is generally not cost effective, as shown by the best- and worst-case analyses. When PCR is assumed to have good test accuracy (the best case) then not only are there the additional diagnostic costs associated with sequential testing but there are higher treatment costs associated with missed cases (false negatives following automated urinalysis) in addition to QALY loss from those missed cases. When PCR is assumed to have a relatively low sensitivity (the worst case) then, comapared with using an automated reagent-strip reading device followed by 24-hour urine collection, more cases will be missed as some true positives with automated urinalysis will then be classified as negative (false negative) by the sequential PCR test. This is in addition to the false negatives following automated urinalysis. Therefore, using PCR as a confirmatory test will have a lower QALY gain than 24-hour urine collection, which would legitimately be considered the gold standard in this worst-case scenario. The conditions for the automated reagent-strip reading device followed by PCR to be cost effective require PCR to have much better test accuracy in women with a positive test result from the automated urinalysis than in the general population of pregnant women with hypertension and for there to be a large cost differential in favour of PCR relative to 24-hour urine collection.

A limitation of the model presented here is the way in which QALYs were estimated. Data on life expectancy from life tables were used and life-time QALYs for neonates and their mothers were discounted assuming they live the rest of their lives in perfect health. Clearly this will tend to over-estimate the discounted lifetime QALY. However, given that most ill health occurs at the end of life, this simplifying assumption will have a relatively small impact on the overall discounted QALY. Furthermore, the estimated QALY gain does not take account of morbidity, a bias that works in the opposite direction to the possible over-estimation of QALYs based on neonatal and maternal mortality.

Conclusion

The cost-effectiveness analysis suggests that the test with better sensitivity will often be the cost-effective option, although specificity can also be an important determinant, especially when test sensitivities are similar. When the automated reagent-strip reading device has higher sensitivity and specificity than PCR, it dominates other options. Conversely, if the characteristics of PCR approach those of a gold standard test, as indicated by Leonos-Miranda et al.,91 then PCR alone dominates.

Given the uncertainty about the differences in test accuracy, the GDG considered that using either PCR alone or an automated reagent-strip reading device followed by 24-hour collection were suitable for estimating proteinuria in a secondary care setting and could be justified on economic grounds. If an automated reagent-strip reading device were used for an initial test, then a 24-hour urine collection should be carried out for women with mild or moderate gestational hypertension and a reading of 1+ or more for proteinuria, based on economic grounds alone.

The GDG recognised that, from a practical point of view, PCR estimation is more convenient for the woman and healthcare professionals in that it provides a quicker result than 24-hour urine protein estimation.