It looks like your analysis program included the continuity correction (since I get 0.017 one tailed using continuity corrected chi-square). You may want to see if you need to use that stricter version, based on the expected cell frequencies. The uncorrected chi-square gives a smaller p value. // I agree that either should be two-tailed (i.e., double the one-tailed p value) since the rate could have gone up or down. If your intervention could not plausibly have increased the rate, you could argue for a one-tailed test. that's judgment.// By the way, each patient is counted exactly once, right? 12 prescriptions means that 12 patients had a prescription? If it could have been 8 patients, some of whom had two prescriptions, the z test for percentages (or the equivalent chi-square) are both inappropriate.