ORDER STATA
## Tables for epidemiologists

- 2 × 2 and 2 × 2 stratified table for longitudinal, cohort study, case–control, and matched case–control data
- Odds ratio, incidence ratio, risk ratio, risk difference, and attributable fraction
- Confidence intervals for the above
- Chi-squared, Fishers’s exact, and Mantel–Haenszel tests
- Tests for homogeneity

- Choice of weights for stratified tables: Mantel–Haenszel, standardized, or user specified
- Exact McNemar test for matched case–control data
- Tabulated odds and odds ratios
- Score test for linear trend

Stata has a suite of tools for dealing with 2 × 2 tables, including stratified tables, known collectively as the epitab features. To calculate appropriate statistics and suppress inappropriate statistics, these features are organized in the same way that epidemiologists conceptualize data.

Stata’s **ir** is used with incidence-rate (incidence
density or person-time) data; point estimates and confidence intervals for
the incidence-rate ratio and difference are calculated, along with the
attributable or prevented fractions for the exposed and total populations.

Stata’s **cs** is used with cohort study data with equal
follow-up time per subject. Risk is then the proportion of subjects who
become cases. Point estimates and confidence intervals for the risk
difference, risk ratio, and (optionally) the odds ratio are calculated,
along with attributable or prevented fractions for the exposed and total
population.

Stata’s **cc** is used with case–control and
cross-sectional data. Point estimates and confidence intervals for the odds
ratio are calculated along with attributable or prevented fractions for the
exposed and total population.

**mcc** is used with matched case–control data. It calculates
McNemar’s chi-squared, point estimates, and confidence intervals for
the difference, ratio, and relative difference of the proportion with the
factor, along with the odds ratio.

All these tools come in two flavors: their normal forms and an “immediate” form. In their normal forms, the commands form counts by summing the dataset in use. In their immediate forms, the data are specified on the command line.

For instance, Boice and Monson (1977 and reprinted in Rothman, Greenland, and Lash 2008, 244) reported on breast cancer cases and person-years of observations for women with tuberculosis who were repeatedly exposed to multiple X-ray fluoroscopies and for those not exposed:

X-ray fluoroscopy
Exposed Unexposed

Breast cancer cases 41 15 Person years 28,010 19,017 |

Using the immediate form of **ir**, you specify the values in the table
following the command:

Exposed Unexposed | Total | ||||

Cases | 41 15 | 56 | |||

Person-time | 28010 19017 | 47027 | |||

Incidence rate | .0014638 .0007888 | .0011908 | |||

Point estimate | [95% Conf. Interval] | ||||

Inc. rate diff. | .000675 | .0000749 .0012751 | |||

Inc. rate ratio | 1.855759 | 1.005684 3.6093 (exact) | |||

Attr. frac. ex. | .4611368 | .0056519 .722938 (exact) | |||

Attr. frac. pop | .337618 | ||||

(midp) Pr(k>=41) = 0.0177 (exact) | |||||

(midp) 2*Pr(k>=41) = 0.0355 (exact) |

The grander **ir** itself can work with individual-level or
aggregate data and also work with stratified data. Rothman, Greenland, and Lash
(2008, 264) report results from Doll and Hill (1966) on age-specific
coronary disease deaths among British male doctors in relation to cigarette
smoking:

Smokers Nonsmokers Age Deaths Person-years Deaths Person-years |

35-44 32 52,407 2 18,790 45-54 104 43,248 12 10,673 55-64 206 28,612 28 5,710 65-74 186 12,663 28 2,585 75-84 102 5,317 31 1,462 |

We have entered these data into Stata:

age smokes deaths pyears | |||

1. | 35-44 1 32 52,407 | ||

2. | 35-44 0 2 18,790 | ||

3. | 45-54 1 104 43,248 | ||

4. | 45-54 0 12 10,673 | ||

5. | 55-64 1 206 28,612 | ||

6. | 55-64 0 28 5,710 | ||

7. | 65-74 1 186 12,663 | ||

8. | 65-74 0 28 2,585 | ||

9. | 75-84 1 102 5,317 | ||

10. | 75-84 0 31 1,462 | ||

We can obtain the Mantel–Haenszel combined estimate of the incidence-rate ratio, along with 90% confidence intervals, by typing

age | IRR [90% Conf. Interval] M-H Weight | |||

35-44 | 5.736638 1.704271 33.61646 1.472169 | (exact) | ||

45-54 | 2.138812 1.274552 3.813282 9.624747 | (exact) | ||

55-64 | 1.46824 1.044915 2.110422 23.34176 | (exact) | ||

65-74 | 1.35606 .9626026 1.953505 23.25315 | (exact) | ||

75-84 | .9047304 .6375194 1.305412 24.31435 | (exact) | ||

Crude | 1.719823 1.437544 2.0688 | (exact) | ||

M-H combined | 1.424682 1.194375 1.699399 | |||

Rothman and Greenland (1998, 264) obtain the standardized incidence-rate
ratio and 90% confidence intervals, weighting each age category by the
population of the exposed group, thus producing the standardized mortality
ratio (SMR). This calculation can be reproduced by specifying
**by(age)** to indicate that the table is stratified, and
**istandard** to specify that we want the internally standardized rate:

age | IRR [90% Conf. Interval] Weight | ||

35-44 | 5.736638 1.704271 33.61646 52407 | (exact) | |

45-54 | 2.138812 1.274552 3.813282 43248 | (exact) | |

55-64 | 1.46824 1.044915 2.110422 28612 | (exact) | |

65-74 | 1.35606 .9626026 1.953505 12663 | (exact) | |

75-84 | .9047304 .6375194 1.305412 5317 | (exact) | |

Crude | 1.719823 1.437544 2.0688 | (exact) | |

I. Standardized | 1.417609 1.186541 1.693676 |

If we want the externally standardized ratio (weights proportional to the
population of the unexposed group), we can substitute **estandard** for
**istandard** above.

- Boice, J. D., and R. R. Monson. 1977.
- Breast cancer in women after repeated fluoroscopic examinations of the chest.
*Journal of the National Cancer Institute*59: 823–832.

- Doll, R., and A. B. Hill. 1966.
- Mortality of British doctors in relation to smoking: observations on coronary thrombosis.
In
*Epidemiological Approaches to the Study of Cancer and Other Chronic Diseases*, ed. W. Haenszel. National Cancer Institute Monograph 19: 205–268.

- Rothman, K. J., S. Greenland, and T. L. Lash. 2008.
*Modern Epidemiology*. 3rd ed (Revised). Philadelphia: Lippincott Williams & Wilkins.