Glossary: reading the numbers
The main text uses everyday language as far as possible, but citing research inevitably brings in a few statistical terms. If something doesn't make sense, look it up in this table; on the upstream online search page, resting the mouse on a word with a dotted underline (tapping it on a phone) also pops up an explanation.
| Term | Meaning |
|---|---|
| All-cause mortality (总死亡率) | Over a period of time, the proportion of a group of people who die, whatever the cause. This book uses it to measure “whether you live long”. In the original research it is called all-cause mortality (全因死亡率), abbreviated ACM |
| HR | Hazard ratio. Over the same period, how fast the group that did something has the event (death, illness), divided by the group that didn't. HR 0.87 means 13% lower than the group that didn't, HR 1.21 means 21% higher |
| RR | Relative risk. The ratio of how likely the two groups are to have the event; the number is read the same way as HR |
| OR | Odds ratio. Also a comparison of two groups, just calculated a little differently from RR: when the thing being compared is rare, it is close to RR; when it is common, it makes the gap look bigger than it really is |
| IRR | Incidence rate ratio. The incidence rate is the proportion of people who newly get the disease over a period; divide one group's by the other's, and read it like RR |
| RaR | Ratio of the number of occurrences. It compares how many times something happened (for example, how many falls), not how many people had it happen; read it like RR |
| Standardized mortality ratio | Abbreviated SMR. How many people in this group actually died, divided by “how many ordinary people of the same age would be expected to die”. 5.86 means the number of deaths is 5.86 times that among people of the same age |
| Risk difference | Subtract one group's probability of the event from the other's to get directly “how many more per thousand people”. A ratio only says how many times over; the risk difference says how many more actual people |
| 95% CI | 95% confidence interval. Any number a study calculates has error; this is the range the true value very likely falls in. For ratio-type numbers, a range that crosses 1, and for difference-type numbers, a range that crosses 0, means the difference between the two groups may be just chance, and the text will say “not statistically significant” |
| RCT | Randomized controlled trial. People are split into two groups by something like drawing lots; one group does the thing and the other doesn't, and afterward you compare how much the results differ. To prove “it was this thing that had the effect”, this method is the most convincing |
| Meta-analysis (荟萃分析) | Pool the results of many studies and recalculate them to get one overall number. Also called meta-analysis (meta 分析) |
| Cohort | Cohort study. Pick a large group of people and follow them for many years, recording who has the event. It can show that two things often occur together, but is not enough to prove the first caused the second |
| Observational | Observational study. The researchers only record from the sidelines, without arranging who does something and who doesn't. Results are easily skewed by other factors (see the confounding and reverse causation entries in this table), so the numbers should be discounted somewhat |
| Confounding | A third factor drives both the cause and the effect, making two things look causally linked. For example, people who like eating vegetables also tend to exercise more, so you can't tell which one is having the effect |
| Reverse causation | The direction of cause and effect is backwards: it looks as if the first thing causes the second, but in fact the second causes the first. It is not that sleeping a lot makes people die early, but that seriously ill people sleep a lot |
| d, g | Effect size, showing how big the gap between two groups is. The gap is converted into “how many times the usual spread” it amounts to: 0.2 counts as small, 0.5 medium, 0.8 large |
| r | Correlation coefficient. How closely two things move together, ranging from -1 to 1; 0.1 counts as weak, 0.3 medium, 0.5 strong |
| MET | A unit of exercise intensity. Sitting quietly is 1 MET; brisk walking is about 3 to 4 MET. Written as MET·h, it is intensity multiplied by hours, showing how much activity in total |
| GRADE | An internationally used scoring system that rates how reliable a piece of evidence is, in four levels: high, moderate, low and very low |
| Intention-to-screen analysis | The effect is calculated by “who was invited to be screened”, not by “who actually got screened”. There are always some among those invited who don't go, so this method underestimates the benefit of screening for those who actually go |
| Pack-years | A unit for how much a person has smoked in total. Packs per day multiplied by years smoked; 30 pack-years means a pack a day for 30 years |
| BMI | Body mass index, a common number for judging how heavy or thin someone is: weight in kilograms divided by the square of height in meters |
| LDL | Low-density lipoprotein cholesterol, commonly called bad cholesterol |
| eGFR | Estimated glomerular filtration rate, a measure of how well the kidneys filter, in mL/min/1.73 m². The lower the number, the worse the kidney function; the stages of chronic kidney disease are set by it |
| HBsAg | Hepatitis B surface antigen, an item on lab reports. A positive result means you are already infected with the hepatitis B virus |
| HPV | Human papillomavirus. It comes in many types, and long-term infection with some of them causes cervical cancer |
| LDCT | Low-dose chest CT, with a radiation dose about one-fifth to one-tenth of a regular CT |
| PM2.5 | Fine particulate matter floating in the air, 2.5 micrometers or less in diameter |
| NOVA | A way of classifying foods, splitting them into four groups by how heavily processed they are; “ultra-processed food” is the fourth group |
| LPR | Loan Prime Rate. The benchmark for loan interest rates in China, published once a month on the 20th; mortgage rates and the cap on interest in private lending are both calculated from it |
| Final arbitration award (一裁终局) | The labor arbitration award takes effect as soon as it is issued, and the employer can no longer take the matter to court |
| Crude marriage rate, crude divorce rate | The number of couples per thousand people who registered a marriage, or a divorce, that year. There is also a figure called the divorce-to-marriage ratio, which is that year's number of divorcing couples divided by the number of marrying couples; it is not the same thing as these two, so don't mix them up |
| AED | Automated external defibrillator. The red or yellow first-aid box common in public places; once it is switched on, just follow the voice prompts, and it decides on its own whether a shock is needed |
| CPR | Cardiopulmonary resuscitation. In cardiac arrest, pressing hard on the chest to keep the blood flowing |
| 3C certification | China Compulsory Certification. Products listed in the catalog may not leave the factory or be sold without this mark |
| ICP filing | A registration procedure in the Ministry of Industry and Information Technology's system before a website or app formally goes live |
| Multi-level protection scheme (等级保护) | The cybersecurity multi-level protection system. Systems are divided into several levels by how important they are; the higher the level, the more security measures are required |
| GPL | An open-source license. When you distribute something built with this kind of code, you usually also have to publish your own code with it |
| Non-compete restriction | An agreement signed with a company: for a period after you leave, you don't go to work for a competitor in the same line of business. During this period the company must pay you compensation every month; it lasts at most 2 years |
| Subscribed capital | The money you promise to put in when registering a company. Once promised, it must actually be paid within the deadline set by law; it is not just a number written down to look good |
| Deposit (定金) vs advance payment (订金) | The two words differ by a single character, but their legal effect differs a lot: a deposit (定金) carries a penalty rule, so if the party that received the money backs out it must refund double, and the amount is at most 20% of the contract amount; an advance payment (订金) is just money paid ahead of time, and backing out carries no such penalty |