Negative Tests and Residual Disease Risk

Why a Negative Result Does Not Always Rule Out Disease

A negative screening result is reassuring only to the extent that the test tends to return negative results for people who do not have the disease, while rarely missing people who do. Even a test with high sensitivity can miss some cases. How much those missed cases matter among all negative results also depends on how common the disease is in the tested population.

The quantity of interest is the residual disease risk: the probability that a person has the disease given that their test result is negative. It is not the same as the probability that a test is negative for someone who has the disease. The latter is the false-negative rate; the former also accounts for negative results among people without the disease.

Definition: Let \(D\) denote the event that a person has the disease, and let \(N\) denote the event that the test result is negative. The residual disease risk after a negative result is \(P(D\mid N)\). The negative predictive value is \(P(\text{not }D\mid N)\), the probability that a person does not have the disease given a negative result. Since \(D\) and not \(D\) are complementary events, the negative predictive value equals \(1-P(D\mid N)\).

From Prevalence and Test Accuracy to Residual Risk

The prevalence \(p\) is the probability that a randomly selected person in the population being tested has the disease. The sensitivity is \(P(\text{positive}\mid D)\), the probability of a positive result among people who have the disease. The specificity is \(P(N\mid \text{not }D)\), the probability of a negative result among people who do not have the disease.

Because sensitivity describes positive results among people with disease, the probability of a negative result among people with disease is its complement, \(1-\text{sensitivity}\). This is the false-negative rate. Specificity, in contrast, directly gives the probability of a negative result among people without disease. Those are the two ways a negative result can arise: a missed case or a true negative.

Result: If prevalence is \(p\), sensitivity is \(Se\), and specificity is \(Sp\), then

$$ P(D\mid N)= \frac{(1-Se)p}{(1-Se)p+Sp(1-p)}. $$ The numerator is the probability of both disease and a negative result. The denominator is the probability of a negative result, including both people with disease and people without disease.

This formula is Bayes’ rule written in terms of the test characteristics. It requires that prevalence and the reported sensitivity and specificity apply to the population and test setting of interest. It also treats disease status as either \(D\) or not \(D\), and the test result as either positive or negative. The supplied values must describe the same relevant population; if test performance differs substantially across groups, one set of sensitivity and specificity values may not be suitable for everyone.

Deriving the Formula

The conditional-probability definition gives \(P(D\mid N)=P(D\cap N)/P(N)\), provided \(P(N)>0\). By the multiplication rule, \(P(D\cap N)=P(N\mid D)P(D)=(1-Se)p\). The law of total probability divides all negative results into two disjoint groups: those from people with disease and those from people without disease.

\begin{aligned} P(N) &=P(N\mid D)P(D)+P(N\mid \text{not }D)P(\text{not }D)\\ &=(1-Se)p+Sp(1-p). \end{aligned}

Substituting these two expressions into the conditional-probability definition produces the stated formula. The denominator must include both kinds of negative result. Dividing only by all people with disease, for example, would give the false-negative rate rather than the probability of disease among people who tested negative.

Worked Calculations

Worked Example: A Negative Result When Prevalence Is 2%

Suppose, hypothetically, a screening program tests 10,000 people from a population with disease prevalence 2%. The test has sensitivity 90% and specificity 95%. Find the probability of disease after a negative result.

Check the setup. Prevalence is \(p=0.02\). Sensitivity is \(Se=0.90\), so the false-negative rate is \(1-0.90=0.10\). Specificity is \(Sp=0.95\). The values are assumed to apply to this tested population, and the result is classified as positive or negative.

Use the formula.

P(D\mid N)= \frac{(1-0.90)(0.02)} {(1-0.90)(0.02)+(0.95)(1-0.02)} = \frac{0.002}{0.002+0.931} = \frac{0.002}{0.933} \approx 0.002144.

The residual disease risk is approximately \(0.002144\), or \(0.2144\%\). In this hypothetical population, about 2.14 of every 1,000 people with a negative result would still have the disease.

Check with counts. Of 10,000 people, 200 have disease and 9,800 do not. At 90% sensitivity, 20 of the 200 people with disease test negative. At 95% specificity, \(0.95(9{,}800)=9{,}310\) people without disease test negative. Thus 9,330 people test negative, of whom 20 have disease:

\frac{20}{20+9{,}310} = \frac{20}{9{,}330} \approx 0.002144 = 0.2144\%.

The negative predictive value is \(9{,}310/9{,}330\approx0.997856\), or \(99.7856\%\). It is the complement of the residual risk, not a different estimate of the same event.

Worked Example: Higher Prevalence Raises Residual Risk

In a second hypothetical setting, prevalence is 10%, sensitivity is 80%, and specificity is 90%. Calculate the disease risk after a negative result.

Check the setup. The prevalence \(p=0.10\) describes the population being screened. Sensitivity \(Se=0.80\) gives a false-negative rate of \(0.20\); specificity is \(Sp=0.90\). Assume these test characteristics apply to this population.

Substitute and calculate.

P(D\mid N)= \frac{(0.20)(0.10)} {(0.20)(0.10)+(0.90)(0.90)} = \frac{0.020}{0.020+0.810} = \frac{0.020}{0.830} \approx 0.02410.

The residual risk is about \(2.410\%\). To check, imagine 10,000 people: 1,000 have disease, and 20% of them, or 200, test negative. Of the 9,000 without disease, 90%, or 8,100, test negative. Therefore \(200/(200+8{,}100)\approx0.02410\). A negative result does not reduce risk to zero because some cases are missed.

Worked Example: High Test Accuracy with Low Prevalence

Suppose a hypothetical disease has prevalence 0.5% in the population being screened. A test has 99% sensitivity and 99% specificity. What is the residual risk after a negative result?

Check the setup. The prevalence is \(p=0.005\). Sensitivity \(Se=0.99\) means a false-negative rate of \(0.01\), and specificity is \(Sp=0.99\). Assume those rates apply to the population of interest.

Calculate.

P(D\mid N)= \frac{(0.01)(0.005)} {(0.01)(0.005)+(0.99)(0.995)} = \frac{0.00005}{0.00005+0.98505} = \frac{0.00005}{0.98510} \approx 0.00005076.

As a percentage, this is about \(0.005076\%\). In a hypothetical group of 100,000 people, 500 have disease and 5 of them test negative. Of the 99,500 people without disease, 98,505 test negative. Hence the risk among negative results is \(5/(5+98{,}505)\approx0.00005076\). Even with both sensitivity and specificity at 99%, the result’s meaning depends on prevalence as well as test performance.

Interpreting Residual Risk Carefully

The result \(P(D\mid N)\) is a conditional probability for people with negative results in the population described by the inputs. It does not say that every person with a negative result has that exact probability in a personal, individualized sense. Nor does it establish that disease is impossible after a negative result. It summarizes the remaining risk under the stated prevalence and test-performance assumptions.

Sensitivity and specificity condition in the opposite direction from residual risk: they start with known disease status and describe the test result. Residual risk starts with a negative result and asks about disease status. The direction matters, so these quantities cannot be substituted for one another. For example, a 90% sensitivity means that 10% of people who have disease test negative; it does not mean that 10% of people with negative results have disease.

A useful way to keep the events straight is to name the denominator before calculating. For residual disease risk, the denominator is all negative results. It includes false negatives and true negatives. The numerator is only the false negatives: people who have disease and receive a negative result. The negative predictive value uses the same denominator but has true negatives as its numerator.

Common mistake: “The sensitivity is 90%, so after a negative test the probability of disease is \(1-0.90=10\%\).” This confuses the false-negative rate \(P(N\mid D)\) with the residual risk \(P(D\mid N)\). The correction is to use Bayes’ rule: combine the false-negative rate with prevalence, then divide by the probability of all negative results, including true negatives.

Natural frequencies are a practical check on the calculation. Choose a convenient hypothetical group size, apply prevalence to find how many have disease, then apply the false-negative rate and specificity to find the two groups with negative results. Dividing the number of false negatives by the total number of negative results should agree with the probability formula. If it does not, check whether the denominator accidentally omitted true negatives or whether a rate was applied to the wrong group.

Identify the inputs.
Record prevalence \(p\), sensitivity \(Se\), and specificity \(Sp\) for the relevant population and test.

Find the false-negative rate.
Calculate \(1-Se\), the probability of a negative result among people with disease.

Compute residual risk.
Use \((1-Se)p\) as the numerator and add the true-negative probability \(Sp(1-p)\) to form the full negative-result denominator.

Interpret in context.
Report \(P(D\mid N)\) as a probability or percentage among negative results, and distinguish it from the negative predictive value.

Check Your Understanding

For each question, keep track of which group a conditional probability starts with and which group forms its denominator.

A test has sensitivity 95%. What is its false-negative rate, and what conditional probability does that rate describe?
Write the formula for \(P(D\mid N)\) using prevalence \(p\), sensitivity \(Se\), and specificity \(Sp\). Identify what each term in its denominator counts probabilistically.
A hypothetical test has prevalence 4%, sensitivity 90%, and specificity 96%. Calculate the residual disease risk after a negative result, as a proportion and as a percentage.
Explain why \(1-\text{sensitivity}\) is generally not the probability of disease among people who test negative.
In a group of 5,000 people, 100 have a disease. A test has 80% sensitivity and 90% specificity. How many false negatives and true negatives are expected, and what is the residual risk among negative results?