SEARCH

SEARCH BY CITATION

Keywords:

  • Crohn's disease;
  • disease index;
  • CDAI;
  • Harvey-Bradshaw

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. ACKNOWLEDGMENTS
  7. References

Background: The Crohn's Disease Activity Index (CDAI) was developed in the 1970s to assess the degree of illness in individuals with Crohn's disease and has since been used widely in clinical trials of the condition. The Harvey-Bradshaw Index (HBI) is a simplification of the CDAI, designed to make data collection and computation easier. It is purported, on the basis of a 0.93 correlation coefficient, to give “essentially the same information.” However, correlation is an incomplete way to assess sameness, and this study aimed to develop a method for predicting CDAI from HBI values, including relevant prediction limits. Materials and Methods: Data used in developing both indexes were combined. Single visits of 224 patients with Crohn's disease were plotted on a scattergram. HBI values seen were integers from 0 through 19. Mean and standard deviation of CDAI were determined for each HBI value that included a sufficient number of patients. Standard deviation of CDAI showed a linear increase with increasing HBI. Therefore, regression of CDAI on HBI was weighted on the inverse of the estimated CDAI standard deviation. Results: Regression predicted a 27-CDAI-unit increase for each HBI unit. Calculated 95% prediction limits were almost straight, diverging lines, bracketing 95% of observations. A table gives central tendency and 95% prediction limits of CDAI for any HBI, as well as key clinical benchmarks. Conclusions: There is a good but far from perfect relationship between CDAI and HBI. CDAI is preferred for clinical trials; HBI is easier to use.

Crohn's disease is an inflammatory disease of the small and/or large bowel of imperfectly understood cause that tends to have a chronic course characterized by remissions and exacerbations and to have only nonspecific, protean manifestations. In the early 1970s, planners of the National Cooperative Crohn's Disease Study, finding no prior satisfactory index for degree of sickness for this disease, gathered appropriate data from which they derived the Crohn's Disease Activity Index (CDAI).1 It has since been proven invaluable for serially quantifying degree of sickness in clinical trials of this disease. It is based on 8 clinical variables, 3 derived from a 1-week patient diary (Table 1). Each independent variable was coded so that 0 corresponds to good health and increasing positive values correspond to greater degrees of sickness. The CDAI was developed using multiple regression through the origin to predict physicians' 4-level global assessments, following which coefficients were standardized and simplified. Investigators' evaluations were that CDAI values <150 generally meant quiescent disease and those >450 generally meant extremely severe disease. Changes in index values from visit to visit correlated well with physicians' assessments of change in patient status. The CDAI served the National Cooperative Crohn's Disease Study well. It provided standardized approaches to data collection, to patient inclusion criteria, and to patient management decisions during study and is a principal response measure.2 The CDAI, the most widely used measure of clinical status in controlled clinical trials of Crohn's disease, has been cited >1000 times, as noted in the Science Citation Index. An expert consensus process initiated by the International Organization for Inflammatory Bowel Disease recently reviewed the use of and positive and negative aspects of various indexes and efficacy end points in Crohn's disease.3 They noted that clinical remissions have most commonly been gauged by a drop of CDAI below 150 and/or by the degree of change in CDAI. Recent recommendations have called for an improvement of at least 100 points.

Table 1. Variables and Coefficients of CDAI and HBI
VariableCDAI Variable De.nitions*CDAI Coef.cient, 1976HBI Coef.cient, 1980
  • *

    *HBI de.nitions are the same or similar.

  • Variables 1 through 3 for CDAI were changed here from 7-day total to average per day, and these coef.cients were adjusted accordingly. HBI uses singleday values for these 3 variables.

1No. of liquid or very soft stools per day141
1-AAdjustment if using diarrhea-control medications30
2Abdominal pain rating (0–3)351
3General well-being rating (0–4)491
4No. of 6 types, other occasional Crohn_s findings201
5Abdominal mass (CDAI = 0, 2, 5; HBI = 1–4)101
6Hematocrit, % decrease from expected6
7Body weight, % decrease from expected1
 Range of index values, these data:  
  Low−270
  High66219

The need for a 7-day patient diary has been a major drawback of the CDAI. It is cumbersome, requires prior Institutional Review Board approval (and often informed consent) if part of a clinical trial, and occasionally is filled in by a patient at one time rather than each day. In addition, the response to rapid clinical changes tends to be buffered by the 7-day collection period. The body weight variable is cumbersome, a minor influence on calculated CDAI values, and at times misleading.

In recognition of these problems and to simplify computation, Harvey and Bradshaw modified the CDAI to create their index (the Harvey-Bradshaw Index [HBI]) by using only a single day's reading for diary entries and excluding 3 variables: body weight, hematocrit, and the use of lomotile or narcotics for diarrhea. In addition, code values are added together rather than summing the products of code values and coefficients (Table 1).4 Definitions of variables otherwise remain essentially the same. The Science Citation Index has recorded >400 citations of HBI. In most reports, HBI was used in place of CDAI, usually in pilot or cohort studies.

Neither the CDAI nor HBI includes any laboratory variables indicative of active inflammation such as C-reactive protein. This omission is considered a deficiency by some investigators.

Harvey and Bradshaw presented a scattergram comparing HBI and CDAI in 112 patients. On the basis of a relatively high correlation coefficient, 0.93, they concluded, “Essentially the same information appears to be given by both methods.”

Bland and Altman5 showed that the correlation coefficient is not a sufficient instrument for deciding whether in such comparisons 1 method of clinical measurement may logically be substituted for another. Their approach for 2 separate methods that measure in the same units, where neither is independent nor dependent, was to plot their difference against their mean and then evaluate the magnitude of differences relative to the practical needs of the measurement.

CDAI and HBI measure in different rather than the same units. Because many clinical reports in Crohn's disease have used one, the other, or both indexes and because CDAI is more comprehensive and more widely used, it appeared worthwhile to examine the relationship between these 2 indexes, examining the magnitude of variation in CDAI values for particular HBI values relative to the practical needs for accuracy. In addition, it appeared that there was need for a convenient method for physicians to convert from HBI to CDAI, particularly in evaluating what happens with their patients relative to what is reported in the literature.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. ACKNOWLEDGMENTS
  7. References

Two sources of data were combined for this analysis. First, values from Harvey and Bradshaw's 112 patients were obtained from their scattergram.4 HBI values could be read exactly. CDAI values were read from a 150% enlargement of their figure. Data ticks were joined by grid lines. Data points could then be read with a maximal error of at most 2, perhaps 3, units. Patient values thus obtained had a correlation coefficient with HBI of 0.93, the same as that calculated by Harvey and Bradshaw.

Second was the database of 112 patients originally used in developing the CDAI.1 The data available at this time contain only 7-day sums rather than individual-day values for the 3 patient diary variables. To simulate a single day, these 7-day sums were divided by 7 and then rounded to the nearest integer. This technique would be expected to result in slightly less spread of values than the choice of a single day but otherwise should be unbiased. More than half of such sums for each of the 3 variables were either 0 or exactly divisible by 7, suggesting uniformity during the 7 days. Frenz et al6 compared prospective and retrospective determinations of CDAI, a comparison primarily reflecting changes in these 3 patient-diary variables. They noted a tendency for patients with high retrospective values to have slightly lower prospective values, but to a large extent, this was related to an increase in the intensity of treatment at the visit separating the 2 estimates. Thus, the present approach is not expected to show appreciable bias related to this phenomenon.

HBI added a fourth alternative to abdominal mass, “definite and tender.” This alternative was assigned if abdominal mass was “definite” and abdominal pain was “severe.” Other minor differences in definitions were ignored, and the simplified computation of HBI was used. This group of patients tended to be slightly sicker than Harvey and Bradshaw's patients, but the relationship between HBI and CDAI was otherwise similar. Thus, in examining the 2 sets of data, it does not appear that possible biases are of sufficient magnitude to negate the value of combining them here.

Because these 2 databases were collected >25 years ago, one may wonder whether the relationship between them in patients has changed over the years. Today, there are greater numbers of effective treatments, but the continued usefulness of CDAI and the face validity of variables that are essentially the same in the 2 indexes indicate that this should not be a problem. Patients now tend to be heavier than before, and relative weight is a variable in CDAI but not in HBI. CDAI assumes that the relationship between percentage weight loss and degree of sickness from Crohn's disease is independent of presickness weight. That presumption continues to appear reasonable, but in any case, body weight is a variable with only minimal influence on final CDAI scores.

Ethical Considerations

Patient data of this analysis are from 2 sources published several years ago,1,4 and patient identification was not accessed or relevant to this investigation. No new data were collected. The Hines VA Institutional Review Board approved the project of which this data analysis is a part.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. ACKNOWLEDGMENTS
  7. References

Data from a single visit for each of these 224 patients with Crohn's disease were plotted as a scattergram in Figure 1, and intermediate computations on that data are shown in Table 2. Eighteen integer values of HBI were seen in this group, with the greatest concentration of patients at HBI of 0 through 4, dwindling to only scattered numbers per value above an HBI of 13. The standard deviation of CDAI values tended to increase with increasing HBI. Some reasons for this are evident. Variation is minimized as values approach a zero floor, a characteristic of 5 of 7 CDAI variables and all HBI variables. The other 2 CDAI variables are open ended at both the top and bottom, but more variation is expected in sicker patients with higher values. Various statistical texts warn of nonhomogeneous variance from regression, but few indicate what to do in such a case. We improvised on the approach of Montgomery and Peck.7 Fourteen HBI values included at least 3 patients each, for a total of 217 of the 224 patients. Means and standard deviations of CDAIs were determined for these HBI values (Table 2). Linear regression gave a moderately good fit in estimating these standard deviations: Predicted standard deviation from group mean CDAI = 26.31 + 3.20 × HBI.

Table 2. Actual and Regression-predicted CDAI Means and Standard Deviations According to the HBI
HBI Groups  
  CDAIRegression Predicts CDAI
HBI Value, XPatients, nMeanSDMean Y1SD Y2
 03617242626
 12757265330
 23186388033
 3191163810736
 4241443513439
 5131473916142
 6141885718846
 7 92365921649
 8 92322624352
 9 72575327055
10102998229758
11113136432462
12 44065335165
13 33993037868
14 236340571
15 126743274
16 149545978
17 048681
18 253851384
19 141454087
20 056790
Total224Regression equations: Y = a + bX
   a26.2726.31
   b27.043.20
thumbnail image

Figure FIGURE 1. Predicting CDAI from HBI weighted on inverse of standard deviation (224 patients).

Download figure to PowerPoint

Regression of CDAI on HBI for the 224 patients, weighted on inverse of estimated standard deviation, is calculated as follows: predicted mean CDAI = 26.27 + 27.04 × HBI.

The 95% prediction limits of CDAI from HBI form a logical display for assessing the suitability of substituting the latter for the former index. To facilitate computation, no error was assumed in predicting standard deviations from group means. Figure 1 includes not only the scattergram but also the resulting regression line and 95% prediction limits. Of 224 observations, 11 (4.9%) fall outside these limits, and outliers are seen at low, medium, and high HBI levels. These are credible results despite the simplifying assumption of no error in standard deviation predictions. Mean CDAI essentially increases by 27 for each unit of HBI increase. This may be compared with the standard deviation of replicate CDAIs when there is no apparent clinical change, previously reported as 45.1

Figure 2 plots selected CDAI values on a cumulative normal probability scale against HBIs. The cumulative normal probability ordinate is preferable here to a linear scale, because the latter would be compressed at its extremes. If the standard deviation of CDAIs had been consistent across all HBI values, the normal probability scale would have resulted in parallel, straight lines. The curvature seen in Figure 2 results because CDAI standard deviations systematically increase with increasing HBI values. Two heavier curves represent a CDAI of 150, previously identified as an upper limit benchmark for quiescent disease, and 450, the lower limit benchmark for severe disease.1 Other curves are separated by CDAI intervals of 100, a degree of change generally accepted as clinically important. Intersection of an HBI vertical line with any CDAI curve is read on the ordinate as percent probability that an actual CDAI at that HBI exceeds the value of the label of the curve. Dotted horizontal lines are shown for selected percentage values. The solid horizontal lines represent the median (50%), a useful central tendency indicator, and the central 95% prediction limits (2.5% and 97.5%).

thumbnail image

Figure FIGURE 2. Predicting CDAI from HBI with 95% prediction limits, cumulative normal probability scale.

Download figure to PowerPoint

As an example, suppose HBI is 8. Figure 2 shows equivalents to be ≈250 for median CDAI and ≈150 to 350 for 95% prediction limits. Table 3 gives actual values for CDAI mean and 95% prediction limits corresponding to specific HBI values as calculated. At HBI = 8, they are 243 for median and 142 to 344 for 95% limits, which is close to the visual approximations from the figure.

Table 3. Mean, 95% Prediction Limits, and Clinical Benchmarks of CDAI as Estimated From HBI
 CDAI   
  • *

    *Standard deviation of replication if no change = 45.

  95% Prediction LimitsClinical CDAI Benchmarks
Harvey-Bradshaw IndexMeanLowerUpperAbsoluteMean Response to PrednisoneMean Change*
 026−27800, Asymptomatic 0, No change2
 153−6112   
 28016145  50, Slight change
 310737178 120, 7 wk 
 413458211  110, Moderate change
 516179244<150, Quiescent  
 6188100277 185, 1 wk175, Marked change
 7216121310   
 8243142344 240, Pretreatment 
 9270162377   
10297183411   
11324203444   
12351223478   
13378243513   
14405263547   
15432282581   
16459302616>450, Severe  
17486321651   
18513339686   
19540358722   
20567376758   
Change per unit27     

Table 3 also shows some key clinical benchmarks. The first randomized, controlled clinical trial in Crohn's disease monitored using the CDAI2 found means of 240 before treatment, 185 after 1 week of prednisone, and 120 after 7 weeks of treatment. Equivalent median values of HBI may be inferred from Figure 2 as ≈8, 6, and 3.5.

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. ACKNOWLEDGMENTS
  7. References

How well does HBI predict CDAI levels? The correlation coefficient of 0.93 between these 2 indexes is relatively high, appearing to indicate a good relationship between them. However, the 95% prediction limits provide a much better gauge of potential interchangeability. These limits, along with clinical interpretations of CDAI levels and changes, indicate that despite the high correlation, CDAI is not nearly as well predicted from HBI as one would like. Additional evidence for the greater reliability of CDAI is the theoretical analysis of Jorgensen et al,8 who noted the signal-to-noise ratio of HBI to be only 6.7, whereas that of CDAI was 17.9, more than 2½ times as great.

The standard deviation from regression of CDAI increases with increasing HBI values, so it is appropriate to calculate diverging prediction limits. Ignoring this sort of phenomenon is not unusual in the medical literature. In this case, it would create an erroneous impression of greater variation at low levels and lesser variation at high levels than truly exist. Only 5% of observations fell outside the diverging 95% prediction limits. These outliers appeared at all levels (low, intermediate, and high), confirming the appropriateness of calculating diverging prediction limits. From a statistical standpoint, it would be preferable to see more subjects at the higher end of both scales. However, on examination of the figures, there is no reason to be uneasy about the extent of the limits shown.

Why is HBI not better at predicting CDAI? Less precision is expected with HBI because it is a simplified version of CDAI. It uses only single-day readings, uses only 5 of 8 CDAI variables, and simply sums variable values rather than applying statistically derived, weighted coefficients.

Another possible reason is the “granularity” of HBI, the fact that coding results in only a limited number of nonnegative integer values, whereas CDAI approximates a continuous distribution. Following 1 reviewer's suggestion, the inverse of the second equation was used on CDAI values of these patients to predict HBI, and then that equation per se was used on these predicted HBI values to repredict CDAI. This was done in 2 ways: acting as if HBI were a continuous variable and restricting estimated HBI to rounded, non-negative integers, including 0, as is actually the case. As may be predicted, the first comparison had a perfect R2 of 1.0000 when relating initial values to final estimates of CDAI. The second had an R2 of 0.9912. The difference, 0.0088, indicates that only a trivial amount, <1%, of variance in estimates of CDAI in this group of subjects is explained by inherent granularity of the HBI scale.

Is part of the lack of precision related to deficiencies in my reading of CDAI levels from Harvey and Bradshaw's Figure 4 or in my method of converting original CDAI1 data to HBIs? As discussed in Materials and Methods, those influences appear to be minimal. I did not have the luxury of working with entirely fresh data, but perhaps others will in the future. Whether HBI or CDAI could be improved by adding other variables such as C-reactive protein is beyond the scope of the present discussion.

Where and why should CDAI be used? Sandborn et al3 indicate that it “is currently the gold standard for evaluation of disease activity [in Crohn's disease].” It has an excellent track record in controlled clinical trials and serves as a standard against which all other clinical indexes of Crohn's disease sickness must continue to be measured. However, there are problems with it. It was designed for prospective studies. That an institutional review committee must approve and patient consent must be obtained for prospective research protocols are hurdles with which researchers must contend. The need for a 7-day patient diary is a major drawback. Ensuring that a patient has a diary card in hand at the appropriate time may be cumbersome to both patient and staff. In some cases, extra visits may be required, 1 to issue the diary and another 1 week later to collect it. Because it incorporates data from 7 consecutive days, CDAI is not sensitive to more rapid changes in clinical status. Each CDAI reading requires a blood hemoglobin determination, one that otherwise may not have been ordered, adding to the cost of care.

The review of monitoring instruments for Crohn's disease by Sandborn et al3 indicates that under certain circumstances additional information on or sensitivity to aspects of a patient's illness and progress not adequately reflected in CDAI may be desired. These aspects include biochemical indicators of inflammation and indicators of enterocutaneous fistula drainage, endoscopic disease activity, histological disease activity, and quality of life.

Sandborn et al3 discuss CDAI levels that have been used for various decisions in clinical trials. They note definitions of <150 as remission, 150 to 219 as mildly active disease, 220 to 450 as moderately active disease, and >450 as severely active disease. Although earlier investigators set the lower limit for admission of patients to study at 150, a more recent trend is to use higher values (e.g., 220). For predominantly inflammatory Crohn's disease, they recommend a primary objective of inducing a clinical remission, CDAI <150, and a secondary objective of decreasing CDAI by at least a given degree (e.g., 70–100 points). They recommend that a subsequent relapse be defined as CDAI ≥150 and an increase by at least 70 points. The bottom line is that although other information should often be collected and analyzed in studies of Crohn's disease, CDAI remains the gold standard of clinical status.

In planning the numbers of subjects needed for a proposed study, an estimate of the initial expected parameters is needed. Assuming that patients are entered into a study with CDAI values of 220 to 450 and that those studied here are typical of what may be expected, the initial mean CDAI is calculated as 256 and the standard deviation as 54.

HBI is good enough and easy enough to use for routine gastroenterology practice. It also may be considered for pilot, cohort, and/or unfunded studies, but it is not recommended for pivotal, prospective trials in Crohn's disease. The cost and bother of using it are clearly less than for the CDAI, but as Jorgensen et al8 and this study have shown, its precision is also less.

I have converted the CDAI recommendations of Sandborn et al3 to corresponding HBI values using Figure 2 and Table 3. Recommendations tend to be conservative, reflecting the fact that HBI is less robust than CDAI. HBI <5 is defined asremission; 5 to 7, as mildly active disease; 8-16, as moderately active disease; and >16, as severely active disease. Although it is reasonable in the practice setting to initiate treatment when HBI is ≥5, it is recommended for comparative studies that only patients with initial HBI ≥8 be included in a therapeutic study. For predominantly inflammatory Crohn's disease, the primary objective of treatment should be induction of clinical remission, HBI <5, and a secondary objective may be a decrease in HBI by at least a given number of points, reasonable values to consider being 3, 4, or 5. Subsequent relapse may be defined as HBI ≥5 and an increase of at least a given number of points, reasonable values to consider being 2, 3, or 4.

In planning the numbers of subjects needed for a proposed study, an estimate of the initial expected parameters is needed. Assuming that patients are entered into a study with HBI values of 8 to 16 and that those studied here are typical of what may be expected, the initial mean HBI is calculated as 10.5 and the standard deviation as 2.0.

Figure 2 and/or Table 3 should be helpful in studying individual patients using HBI and in comparing their courses with published results from clinical trials that used CDAI.

ACKNOWLEDGMENTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. ACKNOWLEDGMENTS
  7. References

The author gratefully acknowledges the Department of Veterans Affairs' continued support of his scholarly activities. He also thanks the reviewers for their critiques and suggestions.

References

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. ACKNOWLEDGMENTS
  7. References
  • 1
    Best WR, Becktel JM, Singleton JW, et al. Development of a Crohn's disease activity index: National Cooperative Crohn's Disease Study. Gastroenterology. 1976; 70:439444.
  • 2
    Summers RW, Switz DM, Sessions JT Jr, et al. National Cooperative Crohn's Disease Study: results of drug treatment. Gastroenterology. 1979; 77:847860.
  • 3
    Sandborn WJ, Feagan BG, Hanauer SB, et al. A review of activity indices and efficacy endpoints for clinical trials of medical therapy in adults with Crohn's disease. Gastroenterology. 2002; 122:512530.
  • 4
    Harvey RF, Bradshaw JM. A simple index of Crohn's disease activity. Lancet. 1980; 1:514.
  • 5
    Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986; 1:307310.
  • 6
    Frenz MB, Dunckley P, Camporota L, et al. Comparison between prospective and retrospective evaluation of Crohn's disease activity index. Am J Gastroenterol. 2005; 100:11171120.
    Direct Link:
  • 7
    Montgomery DC, Peck EA. Weighted least squares. In: Introduction to Linear Regression Analysis. 2nd ed. New York: John Wiley & Sons; 1992:108113.
  • 8
    Jorgensen LG, Fredholm L, Hyltoft Petersen P, et al. How accurate are clinical activity indices for scoring of disease activity in inflammatory bowel disease (IBD)? Clin Chem Lab Med. 2005; 43:403411.