### Friday, February 27, 2004

## Weight Gain and Breast Cancer

Interesting abstract today. Too bad the site doesn't allow full text.

Here is my best shot at NNT (or harm in this case) given the limited information:

1. Start with 62,756 women then have 1,934 cancers so risk = 1934/62756 = 3.01%.

2. "Weight gain of 21-30 pounds was associated with a rate ratio of 1.4%" with an acceptable CI. So, 3.01 x 1.4 = 4.31% and ARR = 4.31 - 3.01 = 1.3%. Therefore, number needed to harm (NNH) = 100 / 1.3 = 77.

3. And "rates doubled among women gaining > 70 pounds" (but no CI information in abstract). So, 3.01 x 2 = 6.02% and ARR = 3.01% and NNH = 100 / 3.01 = 33.

These numbers seem important enough to me to reconsider the traditional risk factors.

Here is my best shot at NNT (or harm in this case) given the limited information:

1. Start with 62,756 women then have 1,934 cancers so risk = 1934/62756 = 3.01%.

2. "Weight gain of 21-30 pounds was associated with a rate ratio of 1.4%" with an acceptable CI. So, 3.01 x 1.4 = 4.31% and ARR = 4.31 - 3.01 = 1.3%. Therefore, number needed to harm (NNH) = 100 / 1.3 = 77.

3. And "rates doubled among women gaining > 70 pounds" (but no CI information in abstract). So, 3.01 x 2 = 6.02% and ARR = 3.01% and NNH = 100 / 3.01 = 33.

These numbers seem important enough to me to reconsider the traditional risk factors.

### Thursday, February 26, 2004

## Don't Eat Your Veggies

Yet another example of relative risk reduction (RRR) hype (Yahoo! News - Ten Grams of Dietary Fiber Cuts Heart Risk - Study). I understand the lay press enthusiasm but the actual article isn't any better.

Let's "do the math":

1. Start with pooled lives = 336,244 and person-years of follow-up = 2,506,581. This means 7.45 years for each person.

2. The number of cardiac events = 5249. So, 5249/336,244 = 1.56%

3. The number of deaths = 2011. So, 2011/336,244 = 0.59%.

4. After "energy-adjusted and measurement error-correction" they report 14% RRR for all coronary events and 27% RRR for coronary deaths. Both are statistically valid by confidence intervals (CI). The magnitude difference between the two isn't expected. I expected coronary event reduction to be larger than death.

5. A 14% RRR from 1.56% = an absolute risk of 1.56% x 0.14 = 0.22% ARR for events.

6. A 27% RRR from 0.59% = an absolute risk or 0.59% x 0.27 = 0.16% ARR for death.

7. Number needed to treat (NNT) with 10 g fiber daily for about 7 years to prevent one event = 100/0.22 = 455.

8. NNT to prevent one death = 100/0.16 = 625.

I'm not impressed.

Also, looking at fiber source reveals no statistical significance for vegetables. It only holds for cereal and fruit.

Let's "do the math":

1. Start with pooled lives = 336,244 and person-years of follow-up = 2,506,581. This means 7.45 years for each person.

2. The number of cardiac events = 5249. So, 5249/336,244 = 1.56%

3. The number of deaths = 2011. So, 2011/336,244 = 0.59%.

4. After "energy-adjusted and measurement error-correction" they report 14% RRR for all coronary events and 27% RRR for coronary deaths. Both are statistically valid by confidence intervals (CI). The magnitude difference between the two isn't expected. I expected coronary event reduction to be larger than death.

5. A 14% RRR from 1.56% = an absolute risk of 1.56% x 0.14 = 0.22% ARR for events.

6. A 27% RRR from 0.59% = an absolute risk or 0.59% x 0.27 = 0.16% ARR for death.

7. Number needed to treat (NNT) with 10 g fiber daily for about 7 years to prevent one event = 100/0.22 = 455.

8. NNT to prevent one death = 100/0.16 = 625.

I'm not impressed.

Also, looking at fiber source reveals no statistical significance for vegetables. It only holds for cereal and fruit.

## Stupid Specialists Tricks #2

I have an elderly male, recovering alcoholic patient. He has osteoporosis on his plain films from his most recent fall and long bone fracture a couple of years ago.

About a year ago his orthopedist recommended DEXA scan testing. I told him it wouldn't change the treatment options and we agreed to treat him with a Rx biphosphonate instead of doing the test.

He was in to see his orthopedist recently because he was disappointed with his surgery outcome. This time the orthopod didn't ask me to arrange the test. He ordered it. When the results were back the ortho told the patient to ask me for an interpretation.

Now it gets interesting. I got a call out of nowhere last week from the patient about the test I did not order (and denied a year ago). I asked my staff to suggest he ask the ordering physician for an interpretation.

Today, the patient was in for an unrelated problem. Yet, most of his visit was spent explaining his recent DEXA scan because he "didn't like the information we gave him". He felt my office wasn't responsive to his needs.

Naturally the DEXA shows osteoporosis.

The patient wanted to know why "40 weeks of Fosamax" had not made his test look better. After suppressing my DOE vs POEM instinct I said "the test was too early". Then, I reminded him, "no one knows the optimal interval for repeat testing and you are already being treated".

Why do the test if it isn't going to change anything?

About a year ago his orthopedist recommended DEXA scan testing. I told him it wouldn't change the treatment options and we agreed to treat him with a Rx biphosphonate instead of doing the test.

He was in to see his orthopedist recently because he was disappointed with his surgery outcome. This time the orthopod didn't ask me to arrange the test. He ordered it. When the results were back the ortho told the patient to ask me for an interpretation.

Now it gets interesting. I got a call out of nowhere last week from the patient about the test I did not order (and denied a year ago). I asked my staff to suggest he ask the ordering physician for an interpretation.

Today, the patient was in for an unrelated problem. Yet, most of his visit was spent explaining his recent DEXA scan because he "didn't like the information we gave him". He felt my office wasn't responsive to his needs.

Naturally the DEXA shows osteoporosis.

The patient wanted to know why "40 weeks of Fosamax" had not made his test look better. After suppressing my DOE vs POEM instinct I said "the test was too early". Then, I reminded him, "no one knows the optimal interval for repeat testing and you are already being treated".

Why do the test if it isn't going to change anything?

### Wednesday, February 04, 2004

## Breast Cancer and Radiotherapy

I recently read with interest A Chance to Cut is a Chance to Cure comments about the benefits of radiotherapy over and above breast conservation therapy for early stage breast cancer.

I find the decision harder when using absolute risk reduction (ARR) instead of relative risk reduction (RRR).

The MSNBC - Radiation after surgery improves survival source is guilty of using RRR to make the benefit seem greater when they say: "women who omitted radiation therapy after surgery were dying at a rate 8.6 percent higher than women who had the radiation".

Here is the citation but you need the full text to go beyond RRR.

This is my math:

1. Rate of events in radiotherapy arm = 755 deaths/4109 patients = 0.1837

2. Rate of events without radiotherapy = 824 deaths/4097 patients = 0.2011

3. Using the UBC Clinical Significance Calculator I get...

4. Absolute risk reduction (ARR) = 0.174

5. Relative risk reduction (RRR) = 9% (same as quoted, but rounded off)

6. Number needed to treat (NNT) = 57. Or, in other words, I would have to recommend this to 57 women to prevent one death.

7. Also, note for 95% confidence NNT ranges between 29 and 2883.

The article does point out "no statistically significant differences in overall survival were found in any individual trial".

The mortality data comes from 13 trials and 8243 randomly assigned patients. If the study was limited just to trials with 1) published data (instead of only the abstract), 2) more than 5 year follow up (because mortality needs relatively long follow-up), and 3) complete data on treatment assignment we find "

It is possible to reach statistical significance without reaching clinical importance.

Also, we have the POEM & DOE problem with recurrence versus death.

The MSNBC article says "A National Institutes of Health panel concluded in 2000 that radiation is necessary for all women who undergo a lumpectomy."

If the rationale for breast conservation surgery over mastectomy (NEJM -- Twenty-Year Follow-up of a Randomized Trial Comparing Total Mastectomy, Lumpectomy, and Lumpectomy plus Irradiation for the Treatment of Invasive Breast Cancer) is higher recurrence but no difference in mortality then why such a strong endorsement for radiotherapy? Isn't it the same argument?

This seems like an opportunity for shared decision making to me.

I find the decision harder when using absolute risk reduction (ARR) instead of relative risk reduction (RRR).

The MSNBC - Radiation after surgery improves survival source is guilty of using RRR to make the benefit seem greater when they say: "women who omitted radiation therapy after surgery were dying at a rate 8.6 percent higher than women who had the radiation".

Here is the citation but you need the full text to go beyond RRR.

This is my math:

1. Rate of events in radiotherapy arm = 755 deaths/4109 patients = 0.1837

2. Rate of events without radiotherapy = 824 deaths/4097 patients = 0.2011

3. Using the UBC Clinical Significance Calculator I get...

4. Absolute risk reduction (ARR) = 0.174

5. Relative risk reduction (RRR) = 9% (same as quoted, but rounded off)

6. Number needed to treat (NNT) = 57. Or, in other words, I would have to recommend this to 57 women to prevent one death.

7. Also, note for 95% confidence NNT ranges between 29 and 2883.

The article does point out "no statistically significant differences in overall survival were found in any individual trial".

The mortality data comes from 13 trials and 8243 randomly assigned patients. If the study was limited just to trials with 1) published data (instead of only the abstract), 2) more than 5 year follow up (because mortality needs relatively long follow-up), and 3) complete data on treatment assignment we find "

*loss of statistical significance*". [My italics]It is possible to reach statistical significance without reaching clinical importance.

Also, we have the POEM & DOE problem with recurrence versus death.

The MSNBC article says "A National Institutes of Health panel concluded in 2000 that radiation is necessary for all women who undergo a lumpectomy."

If the rationale for breast conservation surgery over mastectomy (NEJM -- Twenty-Year Follow-up of a Randomized Trial Comparing Total Mastectomy, Lumpectomy, and Lumpectomy plus Irradiation for the Treatment of Invasive Breast Cancer) is higher recurrence but no difference in mortality then why such a strong endorsement for radiotherapy? Isn't it the same argument?

This seems like an opportunity for shared decision making to me.